[ViewVC] Diff of: jsr166/jsr166/src/jsr166y/ForkJoinPool.java

Comparing jsr166/src/jsr166y/ForkJoinPool.java (file contents):
Revision 1.117 by jsr166, Sat Jan 28 04:32:25 2012 UTC vs.
Revision 1.127 by dl, Sun Mar 4 15:52:45 2012 UTC

#	Line 5 \| Line 5
5		*/
6
7		package jsr166y;
8	–
8		import java.util.ArrayList;
9		import java.util.Arrays;
10		import java.util.Collection;
#	Line 21 \| Line 20 \| import java.util.concurrent.RunnableFutu
20		import java.util.concurrent.TimeUnit;
21		import java.util.concurrent.atomic.AtomicInteger;
22		import java.util.concurrent.atomic.AtomicLong;
23	<	import java.util.concurrent.locks.ReentrantLock;
23	>	import java.util.concurrent.locks.AbstractQueuedSynchronizer;
24		import java.util.concurrent.locks.Condition;
25
26		/**
#	Line 177 \| Line 176 \| public class ForkJoinPool extends Abstra
176		* If an attempted steal fails, a thief always chooses a different
177		* random victim target to try next. So, in order for one thief to
178		* progress, it suffices for any in-progress poll or new push on
179	<	* any empty queue to complete.
179	>	* any empty queue to complete. (This is why we normally use
180	>	* method pollAt and its variants that try once at the apparent
181	>	* base index, else consider alternative actions, rather than
182	>	* method poll.)
183		*
184		* This approach also enables support of a user mode in which local
185		* task processing is in FIFO, not LIFO order, simply by using
#	Line 203 \| Line 205 \| public class ForkJoinPool extends Abstra
205		* take tasks, and they are multiplexed on to a finite number of
206		* shared work queues. However, classes are set up so that future
207		* extensions could allow submitters to optionally help perform
208	<	* tasks as well. Pool submissions from internal workers are also
209	<	* allowed, but use randomized rather than thread-hashed queue
210	<	* indices to avoid imbalance. Insertion of tasks in shared mode
211	<	* requires a lock (mainly to protect in the case of resizing) but
212	<	* we use only a simple spinlock (using bits in field runState),
213	<	* because submitters encountering a busy queue try or create
212	<	* others so never block.
208	>	* tasks as well. Insertion of tasks in shared mode requires a
209	>	* lock (mainly to protect in the case of resizing) but we use
210	>	* only a simple spinlock (using bits in field runState), because
211	>	* submitters encountering a busy queue move on to try or create
212	>	* other queues -- they block only when creating and registering
213	>	* new queues.
214		*
215		* Management
216		* ==========
#	Line 247 \| Line 248 \| public class ForkJoinPool extends Abstra
248		* readers must tolerate null slots. Shared (submission) queues
249		* are at even indices, worker queues at odd indices. Grouping
250		* them together in this way simplifies and speeds up task
251	<	* scanning. To avoid flailing during start-up, the array is
252	<	* presized to hold twice #parallelism workers (which is unlikely
253	<	* to need further resizing during execution). But to avoid
254	<	* dealing with so many null slots, variable runState includes a
254	<	* mask for the nearest power of two that contains all current
255	<	* workers. All worker thread creation is on-demand, triggered by
256	<	* task submissions, replacement of terminated workers, and/or
251	>	* scanning.
252	>	*
253	>	* All worker thread creation is on-demand, triggered by task
254	>	* submissions, replacement of terminated workers, and/or
255		* compensation for blocked workers. However, all other support
256		* code is set up to work with other policies. To ensure that we
257		* do not hold on to worker references that would prevent GC, ALL
#	Line 266 \| Line 264 \| public class ForkJoinPool extends Abstra
264		* both index-check and null-check the IDs. All such accesses
265		* ignore bad IDs by returning out early from what they are doing,
266		* since this can only be associated with termination, in which
267	<	* case it is OK to give up.
268	<	*
269	<	* All uses of the workQueues array check that it is non-null
270	<	* (even if previously non-null). This allows nulling during
271	<	* termination, which is currently not necessary, but remains an
272	<	* option for resource-revocation-based shutdown schemes. It also
275	<	* helps reduce JIT issuance of uncommon-trap code, which tends to
267	>	* case it is OK to give up. All uses of the workQueues array
268	>	* also check that it is non-null (even if previously
269	>	* non-null). This allows nulling during termination, which is
270	>	* currently not necessary, but remains an option for
271	>	* resource-revocation-based shutdown schemes. It also helps
272	>	* reduce JIT issuance of uncommon-trap code, which tends to
273		* unnecessarily complicate control flow in some methods.
274		*
275		* Event Queuing. Unlike HPC work-stealing frameworks, we cannot
#	Line 383 \| Line 380 \| public class ForkJoinPool extends Abstra
380		* (http://portal.acm.org/citation.cfm?id=155354). It differs in
381		* that: (1) We only maintain dependency links across workers upon
382		* steals, rather than use per-task bookkeeping. This sometimes
383	<	* requires a linear scan of workers array to locate stealers, but
384	<	* often doesn't because stealers leave hints (that may become
383	>	* requires a linear scan of workQueues array to locate stealers,
384	>	* but often doesn't because stealers leave hints (that may become
385		* stale/wrong) of where to locate them. A stealHint is only a
386		* hint because a worker might have had multiple steals and the
387		* hint records only one of them (usually the most current).
#	Line 395 \| Line 392 \| public class ForkJoinPool extends Abstra
392		* which means that we miss links in the chain during long-lived
393		* tasks, GC stalls etc (which is OK since blocking in such cases
394		* is usually a good idea). (4) We bound the number of attempts
395	<	* to find work (see MAX_HELP_DEPTH) and fall back to suspending
396	<	* the worker and if necessary replacing it with another.
395	>	* to find work (see MAX_HELP) and fall back to suspending the
396	>	* worker and if necessary replacing it with another.
397		*
398		* It is impossible to keep exactly the target parallelism number
399		* of threads running at any given time. Determining the
400		* existence of conservatively safe helping targets, the
401		* availability of already-created spares, and the apparent need
402		* to create new spares are all racy, so we rely on multiple
403	<	* retries of each. Currently, in keeping with on-demand
404	<	* signalling policy, we compensate only if blocking would leave
405	<	* less than one active (non-waiting, non-blocked) worker.
406	<	* Additionally, to avoid some false alarms due to GC, lagging
407	<	* counters, system activity, etc, compensated blocking for joins
408	<	* is only attempted after rechecks stabilize in
409	<	* ForkJoinTask.awaitJoin. (Retries are interspersed with
410	<	* Thread.yield, for good citizenship.)
403	>	* retries of each. Compensation in the apparent absence of
404	>	* helping opportunities is challenging to control on JVMs, where
405	>	* GC and other activities can stall progress of tasks that in
406	>	* turn stall out many other dependent tasks, without us being
407	>	* able to determine whether they will ever require compensation.
408	>	* Even though work-stealing otherwise encounters little
409	>	* degradation in the presence of more threads than cores,
410	>	* aggressively adding new threads in such cases entails risk of
411	>	* unwanted positive feedback control loops in which more threads
412	>	* cause more dependent stalls (as well as delayed progress of
413	>	* unblocked threads to the point that we know they are available)
414	>	* leading to more situations requiring more threads, and so
415	>	* on. This aspect of control can be seen as an (analytically
416	>	* intractable) game with an opponent that may choose the worst
417	>	* (for us) active thread to stall at any time. We take several
418	>	* precautions to bound losses (and thus bound gains), mainly in
419	>	* methods tryCompensate and awaitJoin: (1) We only try
420	>	* compensation after attempting enough helping steps (measured
421	>	* via counting and timing) that we have already consumed the
422	>	* estimated cost of creating and activating a new thread. (2) We
423	>	* allow up to 50% of threads to be blocked before initially
424	>	* adding any others, and unless completely saturated, check that
425	>	* some work is available for a new worker before adding. Also, we
426	>	* create up to only 50% more threads until entering a mode that
427	>	* only adds a thread if all others are possibly blocked. All
428	>	* together, this means that we might be half as fast to react,
429	>	* and create half as many threads as possible in the ideal case,
430	>	* but present vastly fewer anomalies in all other cases compared
431	>	* to both more aggressive and more conservative alternatives.
432		*
433		* Style notes: There is a lot of representation-level coupling
434		* among classes ForkJoinPool, ForkJoinWorkerThread, and
#	Line 418 \| Line 436 \| public class ForkJoinPool extends Abstra
436		* managed by ForkJoinPool, so are directly accessed. There is
437		* little point trying to reduce this, since any associated future
438		* changes in representations will need to be accompanied by
439	<	* algorithmic changes anyway. All together, these low-level
440	<	* implementation choices produce as much as a factor of 4
441	<	* performance improvement compared to naive implementations, and
442	<	* enable the processing of billions of tasks per second, at the
443	<	* expense of some ugliness.
444	<	*
445	<	* Methods signalWork() and scan() are the main bottlenecks, so are
446	<	* especially heavily micro-optimized/mangled. There are lots of
447	<	* inline assignments (of form "while ((local = field) != 0)")
448	<	* which are usually the simplest way to ensure the required read
449	<	* orderings (which are sometimes critical). This leads to a
450	<	* "C"-like style of listing declarations of these locals at the
451	<	* heads of methods or blocks. There are several occurrences of
452	<	* the unusual "do {} while (!cas...)" which is the simplest way
435	<	* to force an update of a CAS'ed variable. There are also other
436	<	* coding oddities that help some methods perform reasonably even
437	<	* when interpreted (not compiled).
439	>	* algorithmic changes anyway. Several methods intrinsically
440	>	* sprawl because they must accumulate sets of consistent reads of
441	>	* volatiles held in local variables. Methods signalWork() and
442	>	* scan() are the main bottlenecks, so are especially heavily
443	>	* micro-optimized/mangled. There are lots of inline assignments
444	>	* (of form "while ((local = field) != 0)") which are usually the
445	>	* simplest way to ensure the required read orderings (which are
446	>	* sometimes critical). This leads to a "C"-like style of listing
447	>	* declarations of these locals at the heads of methods or blocks.
448	>	* There are several occurrences of the unusual "do {} while
449	>	* (!cas...)" which is the simplest way to force an update of a
450	>	* CAS'ed variable. There are also other coding oddities that help
451	>	* some methods perform reasonably even when interpreted (not
452	>	* compiled).
453		*
454		* The order of declarations in this file is:
455	<	* (1) statics
456	<	* (2) fields (along with constants used when unpacking some of
457	<	* them), listed in an order that tends to reduce contention
458	<	* among them a bit under most JVMs;
459	<	* (3) nested classes
460	<	* (4) internal control methods
461	<	* (5) callbacks and other support for ForkJoinTask methods
462	<	* (6) exported methods (plus a few little helpers)
448	<	* (7) static block initializing all statics in a minimally
449	<	* dependent order.
455	>	* (1) Static utility functions
456	>	* (2) Nested (static) classes
457	>	* (3) Static fields
458	>	* (4) Fields, along with constants used when unpacking some of them
459	>	* (5) Internal control methods
460	>	* (6) Callbacks and other support for ForkJoinTask methods
461	>	* (7) Exported methods
462	>	* (8) Static block initializing statics in minimally dependent order
463		*/
464
465	+	// Static utilities
466	+
467	+	/**
468	+	* If there is a security manager, makes sure caller has
469	+	* permission to modify threads.
470	+	*/
471	+	private static void checkPermission() {
472	+	SecurityManager security = System.getSecurityManager();
473	+	if (security != null)
474	+	security.checkPermission(modifyThreadPermission);
475	+	}
476	+
477	+	// Nested classes
478	+
479		/**
480		* Factory for creating new {@link ForkJoinWorkerThread}s.
481		* A {@code ForkJoinWorkerThreadFactory} must be defined and used
#	Line 477 \| Line 504 \| public class ForkJoinPool extends Abstra
504		}
505
506		/**
507	<	* Creates a new ForkJoinWorkerThread. This factory is used unless
508	<	* overridden in ForkJoinPool constructors.
509	<	*/
510	<	public static final ForkJoinWorkerThreadFactory
511	<	defaultForkJoinWorkerThreadFactory;
512	<
513	<	/**
514	<	* Permission required for callers of methods that may start or
515	<	* kill threads.
516	<	*/
517	<	private static final RuntimePermission modifyThreadPermission;
518	<
519	<	/**
520	<	* If there is a security manager, makes sure caller has
521	<	* permission to modify threads.
522	<	*/
523	<	private static void checkPermission() {
524	<	SecurityManager security = System.getSecurityManager();
525	<	if (security != null)
526	<	security.checkPermission(modifyThreadPermission);
507	>	* A simple non-reentrant lock used for exclusion when managing
508	>	* queues and workers. We use a custom lock so that we can readily
509	>	* probe lock state in constructions that check among alternative
510	>	* actions. The lock is normally only very briefly held, and
511	>	* sometimes treated as a spinlock, but other usages block to
512	>	* reduce overall contention in those cases where locked code
513	>	* bodies perform allocation/resizing.
514	>	*/
515	>	static final class Mutex extends AbstractQueuedSynchronizer {
516	>	public final boolean tryAcquire(int ignore) {
517	>	return compareAndSetState(0, 1);
518	>	}
519	>	public final boolean tryRelease(int ignore) {
520	>	setState(0);
521	>	return true;
522	>	}
523	>	public final void lock() { acquire(0); }
524	>	public final void unlock() { release(0); }
525	>	public final boolean isHeldExclusively() { return getState() == 1; }
526	>	public final Condition newCondition() { return new ConditionObject(); }
527		}
528
529		/**
530	<	* Generator for assigning sequence numbers as pool names.
531	<	*/
532	<	private static final AtomicInteger poolNumberGenerator;
533	<
507	<	/**
508	<	* Bits and masks for control variables
509	<	*
510	<	* Field ctl is a long packed with:
511	<	* AC: Number of active running workers minus target parallelism (16 bits)
512	<	* TC: Number of total workers minus target parallelism (16 bits)
513	<	* ST: true if pool is terminating (1 bit)
514	<	* EC: the wait count of top waiting thread (15 bits)
515	<	* ID: ~(poolIndex >>> 1) of top of Treiber stack of waiters (16 bits)
516	<	*
517	<	* When convenient, we can extract the upper 32 bits of counts and
518	<	* the lower 32 bits of queue state, u = (int)(ctl >>> 32) and e =
519	<	* (int)ctl. The ec field is never accessed alone, but always
520	<	* together with id and st. The offsets of counts by the target
521	<	* parallelism and the positionings of fields makes it possible to
522	<	* perform the most common checks via sign tests of fields: When
523	<	* ac is negative, there are not enough active workers, when tc is
524	<	* negative, there are not enough total workers, when id is
525	<	* negative, there is at least one waiting worker, and when e is
526	<	* negative, the pool is terminating. To deal with these possibly
527	<	* negative fields, we use casts in and out of "short" and/or
528	<	* signed shifts to maintain signedness.
529	<	*
530	<	* When a thread is queued (inactivated), its eventCount field is
531	<	* negative, which is the only way to tell if a worker is
532	<	* prevented from executing tasks, even though it must continue to
533	<	* scan for them to avoid queuing races.
534	<	*
535	<	* Field runState is an int packed with:
536	<	* SHUTDOWN: true if shutdown is enabled (1 bit)
537	<	* SEQ: a sequence number updated upon (de)registering workers (15 bits)
538	<	* MASK: mask (power of 2 - 1) covering all registered poolIndexes (16 bits)
539	<	*
540	<	* The combination of mask and sequence number enables simple
541	<	* consistency checks: Staleness of read-only operations on the
542	<	* workers and queues arrays can be checked by comparing runState
543	<	* before vs after the reads. The low 16 bits (i.e, anding with
544	<	* SMASK) hold the smallest power of two covering all worker
545	<	* indices, minus one. The mask for queues (vs workers) is twice
546	<	* this value plus 1.
547	<	*/
548	<
549	<	// bit positions/shifts for fields
550	<	private static final int AC_SHIFT = 48;
551	<	private static final int TC_SHIFT = 32;
552	<	private static final int ST_SHIFT = 31;
553	<	private static final int EC_SHIFT = 16;
554	<
555	<	// bounds
556	<	private static final int MAX_ID = 0x7fff; // max poolIndex
557	<	private static final int SMASK = 0xffff; // mask short bits
558	<	private static final int SHORT_SIGN = 1 << 15;
559	<	private static final int INT_SIGN = 1 << 31;
560	<
561	<	// masks
562	<	private static final long STOP_BIT = 0x0001L << ST_SHIFT;
563	<	private static final long AC_MASK = ((long)SMASK) << AC_SHIFT;
564	<	private static final long TC_MASK = ((long)SMASK) << TC_SHIFT;
565	<
566	<	// units for incrementing and decrementing
567	<	private static final long TC_UNIT = 1L << TC_SHIFT;
568	<	private static final long AC_UNIT = 1L << AC_SHIFT;
569	<
570	<	// masks and units for dealing with u = (int)(ctl >>> 32)
571	<	private static final int UAC_SHIFT = AC_SHIFT - 32;
572	<	private static final int UTC_SHIFT = TC_SHIFT - 32;
573	<	private static final int UAC_MASK = SMASK << UAC_SHIFT;
574	<	private static final int UTC_MASK = SMASK << UTC_SHIFT;
575	<	private static final int UAC_UNIT = 1 << UAC_SHIFT;
576	<	private static final int UTC_UNIT = 1 << UTC_SHIFT;
577	<
578	<	// masks and units for dealing with e = (int)ctl
579	<	private static final int E_MASK = 0x7fffffff; // no STOP_BIT
580	<	private static final int E_SEQ = 1 << EC_SHIFT;
581	<
582	<	// runState bits
583	<	private static final int SHUTDOWN = 1 << 31;
584	<	private static final int RS_SEQ = 1 << 16;
585	<	private static final int RS_SEQ_MASK = 0x7fff0000;
586	<
587	<	// access mode for WorkQueue
588	<	static final int LIFO_QUEUE = 0;
589	<	static final int FIFO_QUEUE = 1;
590	<	static final int SHARED_QUEUE = -1;
591	<
592	<	/**
593	<	* The wakeup interval (in nanoseconds) for a worker waiting for a
594	<	* task when the pool is quiescent to instead try to shrink the
595	<	* number of workers. The exact value does not matter too
596	<	* much. It must be short enough to release resources during
597	<	* sustained periods of idleness, but not so short that threads
598	<	* are continually re-created.
599	<	*/
600	<	private static final long SHRINK_RATE =
601	<	4L * 1000L * 1000L * 1000L; // 4 seconds
602	<
603	<	/**
604	<	* The timeout value for attempted shrinkage, includes
605	<	* some slop to cope with system timer imprecision.
606	<	*/
607	<	private static final long SHRINK_TIMEOUT = SHRINK_RATE - (SHRINK_RATE / 10);
608	<
609	<	/**
610	<	* The maximum stolen->joining link depth allowed in tryHelpStealer.
611	<	* Depths for legitimate chains are unbounded, but we use a fixed
612	<	* constant to avoid (otherwise unchecked) cycles and to bound
613	<	* staleness of traversal parameters at the expense of sometimes
614	<	* blocking when we could be helping.
615	<	*/
616	<	private static final int MAX_HELP_DEPTH = 16;
617	<
618	<	/*
619	<	* Field layout order in this class tends to matter more than one
620	<	* would like. Runtime layout order is only loosely related to
621	<	* declaration order and may differ across JVMs, but the following
622	<	* empirically works OK on current JVMs.
530	>	* Class for artificial tasks that are used to replace the target
531	>	* of local joins if they are removed from an interior queue slot
532	>	* in WorkQueue.tryRemoveAndExec. We don't need the proxy to
533	>	* actually do anything beyond having a unique identity.
534		*/
535	<
536	<	volatile long ctl; // main pool control
537	<	final int parallelism; // parallelism level
538	<	final int localMode; // per-worker scheduling mode
539	<	int nextPoolIndex; // hint used in registerWorker
540	<	volatile int runState; // shutdown status, seq, and mask
630	<	WorkQueue[] workQueues; // main registry
631	<	final ReentrantLock lock; // for registration
632	<	final Condition termination; // for awaitTermination
633	<	final ForkJoinWorkerThreadFactory factory; // factory for new workers
634	<	final Thread.UncaughtExceptionHandler ueh; // per-worker UEH
635	<	final AtomicLong stealCount; // collect counts when terminated
636	<	final AtomicInteger nextWorkerNumber; // to create worker name string
637	<	final String workerNamePrefix; // Prefix for assigning worker names
535	>	static final class EmptyTask extends ForkJoinTask<Void> {
536	>	EmptyTask() { status = ForkJoinTask.NORMAL; } // force done
537	>	public final Void getRawResult() { return null; }
538	>	public final void setRawResult(Void x) {}
539	>	public final boolean exec() { return true; }
540	>	}
541
542		/**
543		* Queues supporting work-stealing as well as external task
#	Line 685 \| Line 588 \| public class ForkJoinPool extends Abstra
588		* avoiding really bad worst-case access. (Until better JVM
589		* support is in place, this padding is dependent on transient
590		* properties of JVM field layout rules.) We also take care in
591	<	* allocating and sizing and resizing the array. Non-shared queue
591	>	* allocating, sizing and resizing the array. Non-shared queue
592		* arrays are initialized (via method growArray) by workers before
593		* use. Others are allocated on first use.
594		*/
595		static final class WorkQueue {
596		/**
597		* Capacity of work-stealing queue array upon initialization.
598	<	* Must be a power of two; at least 4, but set larger to
599	<	* reduce cacheline sharing among queues.
598	>	* Must be a power of two; at least 4, but should be larger to
599	>	* reduce or eliminate cacheline sharing among queues.
600	>	* Currently, it is much larger, as a partial workaround for
601	>	* the fact that JVMs often place arrays in locations that
602	>	* share GC bookkeeping (especially cardmarks) such that
603	>	* per-write accesses encounter serious memory contention.
604		*/
605	<	static final int INITIAL_QUEUE_CAPACITY = 1 << 8;
605	>	static final int INITIAL_QUEUE_CAPACITY = 1 << 13;
606
607		/**
608		* Maximum size for queue arrays. Must be a power of two less
#	Line 719 \| Line 626 \| public class ForkJoinPool extends Abstra
626		volatile int base; // index of next slot for poll
627		int top; // index of next slot for push
628		ForkJoinTask<?>[] array; // the elements (initially unallocated)
629	+	final ForkJoinPool pool; // the containing pool (may be null)
630		final ForkJoinWorkerThread owner; // owning thread or null if shared
631		volatile Thread parker; // == owner during call to park; else null
632		ForkJoinTask<?> currentJoin; // task being joined in awaitJoin
633		ForkJoinTask<?> currentSteal; // current non-local task being executed
634		// Heuristic padding to ameliorate unfortunate memory placements
635	<	Object p00, p01, p02, p03, p04, p05, p06, p07, p08, p09, p0a;
635	>	Object p00, p01, p02, p03, p04, p05, p06, p07;
636	>	Object p08, p09, p0a, p0b, p0c, p0d, p0e;
637
638	<	WorkQueue(ForkJoinWorkerThread owner, int mode) {
730	<	this.owner = owner;
638	>	WorkQueue(ForkJoinPool pool, ForkJoinWorkerThread owner, int mode) {
639		this.mode = mode;
640	+	this.pool = pool;
641	+	this.owner = owner;
642		// Place indices in the center of array (that is not yet allocated)
643		base = top = INITIAL_QUEUE_CAPACITY >>> 1;
644		}
645
646		/**
647	<	* Returns number of tasks in the queue.
647	>	* Returns the approximate number of tasks in the queue.
648		*/
649		final int queueSize() {
650	<	int n = base - top; // non-owner callers must read base first
651	<	return (n >= 0) ? 0 : -n;
650	>	int n = base - top; // non-owner callers must read base first
651	>	return (n >= 0) ? 0 : -n; // ignore transient negative
652	>	}
653	>
654	>	/**
655	>	* Provides a more accurate estimate of whether this queue has
656	>	* any tasks than does queueSize, by checking whether a
657	>	* near-empty queue has at least one unclaimed task.
658	>	*/
659	>	final boolean isEmpty() {
660	>	ForkJoinTask<?>[] a; int m, s;
661	>	int n = base - (s = top);
662	>	return (n >= 0 \|\|
663	>	(n == -1 &&
664	>	((a = array) == null \|\|
665	>	(m = a.length - 1) < 0 \|\|
666	>	U.getObjectVolatile
667	>	(a, ((m & (s - 1)) << ASHIFT) + ABASE) == null)));
668		}
669
670		/**
671		* Pushes a task. Call only by owner in unshared queues.
672		*
673		* @param task the task. Caller must ensure non-null.
748	–	* @param p if non-null, pool to signal if necessary
674		* @throw RejectedExecutionException if array cannot be resized
675		*/
676	<	final void push(ForkJoinTask<?> task, ForkJoinPool p) {
677	<	ForkJoinTask<?>[] a;
676	>	final void push(ForkJoinTask<?> task) {
677	>	ForkJoinTask<?>[] a; ForkJoinPool p;
678		int s = top, m, n;
679		if ((a = array) != null) { // ignore if queue removed
680		U.putOrderedObject
681		(a, (((m = a.length - 1) & s) << ASHIFT) + ABASE, task);
682		if ((n = (top = s + 1) - base) <= 2) {
683	<	if (p != null)
683	>	if ((p = pool) != null)
684		p.signalWork();
685		}
686		else if (n >= m)
#	Line 774 \| Line 699 \| public class ForkJoinPool extends Abstra
699		boolean submitted = false;
700		if (runState == 0 && U.compareAndSwapInt(this, RUNSTATE, 0, 1)) {
701		ForkJoinTask<?>[] a = array;
702	<	int s = top, n = s - base;
702	>	int s = top;
703		try {
704	<	if ((a != null && n < a.length - 1) \|\|
704	>	if ((a != null && a.length > s + 1 - base) \|\|
705		(a = growArray(false)) != null) { // must presize
706		int j = (((a.length - 1) & s) << ASHIFT) + ABASE;
707		U.putObject(a, (long)j, task); // don't need "ordered"
#	Line 791 \| Line 716 \| public class ForkJoinPool extends Abstra
716		}
717
718		/**
719	<	* Takes next task, if one exists, in FIFO order.
719	>	* Takes next task, if one exists, in LIFO order. Call only
720	>	* by owner in unshared queues. (We do not have a shared
721	>	* version of this method because it is never needed.)
722		*/
723	<	final ForkJoinTask<?> poll() {
724	<	ForkJoinTask<?>[] a; int b, i;
725	<	while ((b = base) - top < 0 && (a = array) != null &&
726	<	(i = (a.length - 1) & b) >= 0) {
727	<	int j = (i << ASHIFT) + ABASE;
728	<	ForkJoinTask<?> t = (ForkJoinTask<?>)U.getObjectVolatile(a, j);
729	<	if (t != null && base == b &&
723	>	final ForkJoinTask<?> pop() {
724	>	ForkJoinTask<?>[] a; ForkJoinTask<?> t; int m;
725	>	if ((a = array) != null && (m = a.length - 1) >= 0) {
726	>	for (int s; (s = top - 1) - base >= 0;) {
727	>	long j = ((m & s) << ASHIFT) + ABASE;
728	>	if ((t = (ForkJoinTask<?>)U.getObject(a, j)) == null)
729	>	break;
730	>	if (U.compareAndSwapObject(a, j, t, null)) {
731	>	top = s;
732	>	return t;
733	>	}
734	>	}
735	>	}
736	>	return null;
737	>	}
738	>
739	>	/**
740	>	* Takes a task in FIFO order if b is base of queue and a task
741	>	* can be claimed without contention. Specialized versions
742	>	* appear in ForkJoinPool methods scan and tryHelpStealer.
743	>	*/
744	>	final ForkJoinTask<?> pollAt(int b) {
745	>	ForkJoinTask<?> t; ForkJoinTask<?>[] a;
746	>	if ((a = array) != null) {
747	>	int j = (((a.length - 1) & b) << ASHIFT) + ABASE;
748	>	if ((t = (ForkJoinTask<?>)U.getObjectVolatile(a, j)) != null &&
749	>	base == b &&
750		U.compareAndSwapObject(a, j, t, null)) {
751		base = b + 1;
752		return t;
#	Line 809 \| Line 756 \| public class ForkJoinPool extends Abstra
756		}
757
758		/**
759	<	* Takes next task, if one exists, in LIFO order.
813	<	* Call only by owner in unshared queues.
759	>	* Takes next task, if one exists, in FIFO order.
760		*/
761	<	final ForkJoinTask<?> pop() {
762	<	ForkJoinTask<?> t; int m;
763	<	ForkJoinTask<?>[] a = array;
764	<	if (a != null && (m = a.length - 1) >= 0) {
765	<	for (int s; (s = top - 1) - base >= 0;) {
766	<	int j = ((m & s) << ASHIFT) + ABASE;
767	<	if ((t = (ForkJoinTask<?>)U.getObjectVolatile(a, j)) == null)
768	<	break;
769	<	if (U.compareAndSwapObject(a, j, t, null)) {
824	<	top = s;
761	>	final ForkJoinTask<?> poll() {
762	>	ForkJoinTask<?>[] a; int b; ForkJoinTask<?> t;
763	>	while ((b = base) - top < 0 && (a = array) != null) {
764	>	int j = (((a.length - 1) & b) << ASHIFT) + ABASE;
765	>	t = (ForkJoinTask<?>)U.getObjectVolatile(a, j);
766	>	if (t != null) {
767	>	if (base == b &&
768	>	U.compareAndSwapObject(a, j, t, null)) {
769	>	base = b + 1;
770		return t;
771		}
772		}
773	+	else if (base == b) {
774	+	if (b + 1 == top)
775	+	break;
776	+	Thread.yield(); // wait for lagging update
777	+	}
778		}
779		return null;
780		}
#	Line 849 \| Line 799 \| public class ForkJoinPool extends Abstra
799		}
800
801		/**
852	–	* Returns task at index b if b is current base of queue.
853	–	*/
854	–	final ForkJoinTask<?> pollAt(int b) {
855	–	ForkJoinTask<?>[] a; int i;
856	–	ForkJoinTask<?> task = null;
857	–	if ((a = array) != null && (i = ((a.length - 1) & b)) >= 0) {
858	–	int j = (i << ASHIFT) + ABASE;
859	–	ForkJoinTask<?> t = (ForkJoinTask<?>)U.getObjectVolatile(a, j);
860	–	if (t != null && base == b &&
861	–	U.compareAndSwapObject(a, j, t, null)) {
862	–	base = b + 1;
863	–	task = t;
864	–	}
865	–	}
866	–	return task;
867	–	}
868	–
869	–	/**
802		* Pops the given task only if it is at the current top.
803		*/
804		final boolean tryUnpush(ForkJoinTask<?> t) {
#	Line 884 \| Line 816 \| public class ForkJoinPool extends Abstra
816		* Polls the given task only if it is at the current base.
817		*/
818		final boolean pollFor(ForkJoinTask<?> task) {
819	<	ForkJoinTask<?>[] a; int b, i;
820	<	if ((b = base) - top < 0 && (a = array) != null &&
821	<	(i = (a.length - 1) & b) >= 0) {
890	<	int j = (i << ASHIFT) + ABASE;
819	>	ForkJoinTask<?>[] a; int b;
820	>	if ((b = base) - top < 0 && (a = array) != null) {
821	>	int j = (((a.length - 1) & b) << ASHIFT) + ABASE;
822		if (U.getObjectVolatile(a, j) == task && base == b &&
823		U.compareAndSwapObject(a, j, task, null)) {
824		base = b + 1;
#	Line 898 \| Line 829 \| public class ForkJoinPool extends Abstra
829		}
830
831		/**
901	–	* If present, removes from queue and executes the given task, or
902	–	* any other cancelled task. Returns (true) immediately on any CAS
903	–	* or consistency check failure so caller can retry.
904	–	*
905	–	* @return false if no progress can be made
906	–	*/
907	–	final boolean tryRemoveAndExec(ForkJoinTask<?> task) {
908	–	boolean removed = false, empty = true, progress = true;
909	–	ForkJoinTask<?>[] a; int m, s, b, n;
910	–	if ((a = array) != null && (m = a.length - 1) >= 0 &&
911	–	(n = (s = top) - (b = base)) > 0) {
912	–	for (ForkJoinTask<?> t;;) { // traverse from s to b
913	–	int j = ((--s & m) << ASHIFT) + ABASE;
914	–	t = (ForkJoinTask<?>)U.getObjectVolatile(a, j);
915	–	if (t == null) // inconsistent length
916	–	break;
917	–	else if (t == task) {
918	–	if (s + 1 == top) { // pop
919	–	if (!U.compareAndSwapObject(a, j, task, null))
920	–	break;
921	–	top = s;
922	–	removed = true;
923	–	}
924	–	else if (base == b) // replace with proxy
925	–	removed = U.compareAndSwapObject(a, j, task,
926	–	new EmptyTask());
927	–	break;
928	–	}
929	–	else if (t.status >= 0)
930	–	empty = false;
931	–	else if (s + 1 == top) { // pop and throw away
932	–	if (U.compareAndSwapObject(a, j, t, null))
933	–	top = s;
934	–	break;
935	–	}
936	–	if (--n == 0) {
937	–	if (!empty && base == b)
938	–	progress = false;
939	–	break;
940	–	}
941	–	}
942	–	}
943	–	if (removed)
944	–	task.doExec();
945	–	return progress;
946	–	}
947	–
948	–	/**
832		* Initializes or doubles the capacity of array. Call either
833		* by owner or with lock held -- it is OK for base, but not
834		* top, to move while resizings are in progress.
#	Line 990 \| Line 873 \| public class ForkJoinPool extends Abstra
873		ForkJoinTask.cancelIgnoringExceptions(t);
874		}
875
876	+	/**
877	+	* Computes next value for random probes. Scans don't require
878	+	* a very high quality generator, but also not a crummy one.
879	+	* Marsaglia xor-shift is cheap and works well enough. Note:
880	+	* This is manually inlined in its usages in ForkJoinPool to
881	+	* avoid writes inside busy scan loops.
882	+	*/
883	+	final int nextSeed() {
884	+	int r = seed;
885	+	r ^= r << 13;
886	+	r ^= r >>> 17;
887	+	return seed = r ^= r << 5;
888	+	}
889	+
890		// Execution methods
891
892		/**
893	<	* Removes and runs tasks until empty, using local mode
997	<	* ordering.
893	>	* Pops and runs tasks until empty.
894		*/
895	<	final void runLocalTasks() {
896	<	if (base - top < 0) {
897	<	for (ForkJoinTask<?> t; (t = nextLocalTask()) != null; )
895	>	private void popAndExecAll() {
896	>	// A bit faster than repeated pop calls
897	>	ForkJoinTask<?>[] a; int m, s; long j; ForkJoinTask<?> t;
898	>	while ((a = array) != null && (m = a.length - 1) >= 0 &&
899	>	(s = top - 1) - base >= 0 &&
900	>	(t = ((ForkJoinTask<?>)
901	>	U.getObject(a, j = ((m & s) << ASHIFT) + ABASE)))
902	>	!= null) {
903	>	if (U.compareAndSwapObject(a, j, t, null)) {
904	>	top = s;
905		t.doExec();
906	+	}
907	+	}
908	+	}
909	+
910	+	/**
911	+	* Polls and runs tasks until empty.
912	+	*/
913	+	private void pollAndExecAll() {
914	+	for (ForkJoinTask<?> t; (t = poll()) != null;)
915	+	t.doExec();
916	+	}
917	+
918	+	/**
919	+	* If present, removes from queue and executes the given task, or
920	+	* any other cancelled task. Returns (true) immediately on any CAS
921	+	* or consistency check failure so caller can retry.
922	+	*
923	+	* @return false if no progress can be made
924	+	*/
925	+	final boolean tryRemoveAndExec(ForkJoinTask<?> task) {
926	+	boolean removed = false, empty = true, progress = true;
927	+	ForkJoinTask<?>[] a; int m, s, b, n;
928	+	if ((a = array) != null && (m = a.length - 1) >= 0 &&
929	+	(n = (s = top) - (b = base)) > 0) {
930	+	for (ForkJoinTask<?> t;;) { // traverse from s to b
931	+	int j = ((--s & m) << ASHIFT) + ABASE;
932	+	t = (ForkJoinTask<?>)U.getObjectVolatile(a, j);
933	+	if (t == null) // inconsistent length
934	+	break;
935	+	else if (t == task) {
936	+	if (s + 1 == top) { // pop
937	+	if (!U.compareAndSwapObject(a, j, task, null))
938	+	break;
939	+	top = s;
940	+	removed = true;
941	+	}
942	+	else if (base == b) // replace with proxy
943	+	removed = U.compareAndSwapObject(a, j, task,
944	+	new EmptyTask());
945	+	break;
946	+	}
947	+	else if (t.status >= 0)
948	+	empty = false;
949	+	else if (s + 1 == top) { // pop and throw away
950	+	if (U.compareAndSwapObject(a, j, t, null))
951	+	top = s;
952	+	break;
953	+	}
954	+	if (--n == 0) {
955	+	if (!empty && base == b)
956	+	progress = false;
957	+	break;
958	+	}
959	+	}
960		}
961	+	if (removed)
962	+	task.doExec();
963	+	return progress;
964		}
965
966		/**
967		* Executes a top-level task and any local tasks remaining
968		* after execution.
1009	–	*
1010	–	* @return true unless terminating
969		*/
970	<	final boolean runTask(ForkJoinTask<?> t) {
1013	<	boolean alive = true;
970	>	final void runTask(ForkJoinTask<?> t) {
971		if (t != null) {
972		currentSteal = t;
973		t.doExec();
974	<	runLocalTasks();
974	>	if (top != base) { // process remaining local tasks
975	>	if (mode == 0)
976	>	popAndExecAll();
977	>	else
978	>	pollAndExecAll();
979	>	}
980		++nsteals;
981		currentSteal = null;
982		}
1021	–	else if (runState < 0) // terminating
1022	–	alive = false;
1023	–	return alive;
983		}
984
985		/**
#	Line 1036 \| Line 995 \| public class ForkJoinPool extends Abstra
995		}
996
997		/**
998	<	* Computes next value for random probes. Scans don't require
1040	<	* a very high quality generator, but also not a crummy one.
1041	<	* Marsaglia xor-shift is cheap and works well enough. Note:
1042	<	* This is manually inlined in several usages in ForkJoinPool
1043	<	* to avoid writes inside busy scan loops.
998	>	* Returns true if owned and not known to be blocked.
999		*/
1000	<	final int nextSeed() {
1001	<	int r = seed;
1002	<	r ^= r << 13;
1003	<	r ^= r >>> 17;
1004	<	r ^= r << 5;
1005	<	return seed = r;
1000	>	final boolean isApparentlyUnblocked() {
1001	>	Thread wt; Thread.State s;
1002	>	return (eventCount >= 0 &&
1003	>	(wt = owner) != null &&
1004	>	(s = wt.getState()) != Thread.State.BLOCKED &&
1005	>	s != Thread.State.WAITING &&
1006	>	s != Thread.State.TIMED_WAITING);
1007	>	}
1008	>
1009	>	/**
1010	>	* If this owned and is not already interrupted, try to
1011	>	* interrupt and/or unpark, ignoring exceptions.
1012	>	*/
1013	>	final void interruptOwner() {
1014	>	Thread wt, p;
1015	>	if ((wt = owner) != null && !wt.isInterrupted()) {
1016	>	try {
1017	>	wt.interrupt();
1018	>	} catch (SecurityException ignore) {
1019	>	}
1020	>	}
1021	>	if ((p = parker) != null)
1022	>	U.unpark(p);
1023		}
1024
1025		// Unsafe mechanics
#	Line 1075 \| Line 1047 \| public class ForkJoinPool extends Abstra
1047		}
1048
1049		/**
1050	<	* Class for artificial tasks that are used to replace the target
1051	<	* of local joins if they are removed from an interior queue slot
1052	<	* in WorkQueue.tryRemoveAndExec. We don't need the proxy to
1053	<	* actually do anything beyond having a unique identity.
1054	<	*/
1055	<	static final class EmptyTask extends ForkJoinTask<Void> {
1056	<	EmptyTask() { status = ForkJoinTask.NORMAL; } // force done
1057	<	public Void getRawResult() { return null; }
1058	<	public void setRawResult(Void x) {}
1059	<	public boolean exec() { return true; }
1060	<	}
1061	<
1062	<	/**
1063	<	* Per-thread records for (typically non-FJ) threads that submit
1092	<	* to pools. Cureently holds only psuedo-random seed / index that
1093	<	* is used to choose submission queues in method doSubmit. In the
1094	<	* future, this may incorporate a means to implement different
1095	<	* task rejection and resubmission policies.
1050	>	* Per-thread records for threads that submit to pools. Currently
1051	>	* holds only pseudo-random seed / index that is used to choose
1052	>	* submission queues in method doSubmit. In the future, this may
1053	>	* also incorporate a means to implement different task rejection
1054	>	* and resubmission policies.
1055	>	*
1056	>	* Seeds for submitters and workers/workQueues work in basically
1057	>	* the same way but are initialized and updated using slightly
1058	>	* different mechanics. Both are initialized using the same
1059	>	* approach as in class ThreadLocal, where successive values are
1060	>	* unlikely to collide with previous values. This is done during
1061	>	* registration for workers, but requires a separate AtomicInteger
1062	>	* for submitters. Seeds are then randomly modified upon
1063	>	* collisions using xorshifts, which requires a non-zero seed.
1064		*/
1065		static final class Submitter {
1066	<	int seed; // seed for random submission queue selection
1099	<
1100	<	// Heuristic padding to ameliorate unfortunate memory placements
1101	<	int p0, p1, p2, p3, p4, p5, p6, p7, p8, p9, pa, pb, pc, pd, pe;
1102	<
1066	>	int seed;
1067		Submitter() {
1068	<	// Use identityHashCode, forced negative, for seed
1069	<	seed = System.identityHashCode(Thread.currentThread()) \| (1 << 31);
1106	<	}
1107	<
1108	<	/**
1109	<	* Computes next value for random probes. Like method
1110	<	* WorkQueue.nextSeed, this is manually inlined in several
1111	<	* usages to avoid writes inside busy loops.
1112	<	*/
1113	<	final int nextSeed() {
1114	<	int r = seed;
1115	<	r ^= r << 13;
1116	<	r ^= r >>> 17;
1117	<	return seed = r ^= r << 5;
1068	>	int s = nextSubmitterSeed.getAndAdd(SEED_INCREMENT);
1069	>	seed = (s == 0) ? 1 : s; // ensure non-zero
1070		}
1071		}
1072
#	Line 1123 \| Line 1075 \| public class ForkJoinPool extends Abstra
1075		public Submitter initialValue() { return new Submitter(); }
1076		}
1077
1078	+	// static fields (initialized in static initializer below)
1079	+
1080	+	/**
1081	+	* Creates a new ForkJoinWorkerThread. This factory is used unless
1082	+	* overridden in ForkJoinPool constructors.
1083	+	*/
1084	+	public static final ForkJoinWorkerThreadFactory
1085	+	defaultForkJoinWorkerThreadFactory;
1086	+
1087	+	/**
1088	+	* Generator for assigning sequence numbers as pool names.
1089	+	*/
1090	+	private static final AtomicInteger poolNumberGenerator;
1091	+
1092	+	/**
1093	+	* Generator for initial hashes/seeds for submitters. Accessed by
1094	+	* Submitter class constructor.
1095	+	*/
1096	+	static final AtomicInteger nextSubmitterSeed;
1097	+
1098	+	/**
1099	+	* Permission required for callers of methods that may start or
1100	+	* kill threads.
1101	+	*/
1102	+	private static final RuntimePermission modifyThreadPermission;
1103	+
1104		/**
1105		* Per-thread submission bookeeping. Shared across all pools
1106		* to reduce ThreadLocal pollution and because random motion
1107		* to avoid contention in one pool is likely to hold for others.
1108		*/
1109	<	static final ThreadSubmitter submitters = new ThreadSubmitter();
1109	>	private static final ThreadSubmitter submitters;
1110	>
1111	>	// static constants
1112
1113		/**
1114	<	* Top-level runloop for workers
1114	>	* The wakeup interval (in nanoseconds) for a worker waiting for a
1115	>	* task when the pool is quiescent to instead try to shrink the
1116	>	* number of workers. The exact value does not matter too
1117	>	* much. It must be short enough to release resources during
1118	>	* sustained periods of idleness, but not so short that threads
1119	>	* are continually re-created.
1120		*/
1121	<	final void runWorker(ForkJoinWorkerThread wt) {
1122	<	// Initialize queue array and seed in this thread
1138	<	WorkQueue w = wt.workQueue;
1139	<	w.growArray(false);
1140	<	// Same initial hash as Submitters
1141	<	w.seed = System.identityHashCode(Thread.currentThread()) \| (1 << 31);
1121	>	private static final long SHRINK_RATE =
1122	>	4L * 1000L * 1000L * 1000L; // 4 seconds
1123
1124	<	do {} while (w.runTask(scan(w)));
1125	<	}
1124	>	/**
1125	>	* The timeout value for attempted shrinkage, includes
1126	>	* some slop to cope with system timer imprecision.
1127	>	*/
1128	>	private static final long SHRINK_TIMEOUT = SHRINK_RATE - (SHRINK_RATE / 10);
1129	>
1130	>	/**
1131	>	* The maximum stolen->joining link depth allowed in method
1132	>	* tryHelpStealer. Must be a power of two. This value also
1133	>	* controls the maximum number of times to try to help join a task
1134	>	* without any apparent progress or change in pool state before
1135	>	* giving up and blocking (see awaitJoin). Depths for legitimate
1136	>	* chains are unbounded, but we use a fixed constant to avoid
1137	>	* (otherwise unchecked) cycles and to bound staleness of
1138	>	* traversal parameters at the expense of sometimes blocking when
1139	>	* we could be helping.
1140	>	*/
1141	>	private static final int MAX_HELP = 32;
1142
1143	<	// Creating, registering and deregistering workers
1143	>	/**
1144	>	* Secondary time-based bound (in nanosecs) for helping attempts
1145	>	* before trying compensated blocking in awaitJoin. Used in
1146	>	* conjunction with MAX_HELP to reduce variance due to different
1147	>	* polling rates associated with different helping options. The
1148	>	* value should roughly approximate the time required to create
1149	>	* and/or activate a worker thread.
1150	>	*/
1151	>	private static final long COMPENSATION_DELAY = 100L * 1000L; // 0.1 millisec
1152	>
1153	>	/**
1154	>	* Increment for seed generators. See class ThreadLocal for
1155	>	* explanation.
1156	>	*/
1157	>	private static final int SEED_INCREMENT = 0x61c88647;
1158	>
1159	>	/**
1160	>	* Bits and masks for control variables
1161	>	*
1162	>	* Field ctl is a long packed with:
1163	>	* AC: Number of active running workers minus target parallelism (16 bits)
1164	>	* TC: Number of total workers minus target parallelism (16 bits)
1165	>	* ST: true if pool is terminating (1 bit)
1166	>	* EC: the wait count of top waiting thread (15 bits)
1167	>	* ID: poolIndex of top of Treiber stack of waiters (16 bits)
1168	>	*
1169	>	* When convenient, we can extract the upper 32 bits of counts and
1170	>	* the lower 32 bits of queue state, u = (int)(ctl >>> 32) and e =
1171	>	* (int)ctl. The ec field is never accessed alone, but always
1172	>	* together with id and st. The offsets of counts by the target
1173	>	* parallelism and the positionings of fields makes it possible to
1174	>	* perform the most common checks via sign tests of fields: When
1175	>	* ac is negative, there are not enough active workers, when tc is
1176	>	* negative, there are not enough total workers, and when e is
1177	>	* negative, the pool is terminating. To deal with these possibly
1178	>	* negative fields, we use casts in and out of "short" and/or
1179	>	* signed shifts to maintain signedness.
1180	>	*
1181	>	* When a thread is queued (inactivated), its eventCount field is
1182	>	* set negative, which is the only way to tell if a worker is
1183	>	* prevented from executing tasks, even though it must continue to
1184	>	* scan for them to avoid queuing races. Note however that
1185	>	* eventCount updates lag releases so usage requires care.
1186	>	*
1187	>	* Field runState is an int packed with:
1188	>	* SHUTDOWN: true if shutdown is enabled (1 bit)
1189	>	* SEQ: a sequence number updated upon (de)registering workers (30 bits)
1190	>	* INIT: set true after workQueues array construction (1 bit)
1191	>	*
1192	>	* The sequence number enables simple consistency checks:
1193	>	* Staleness of read-only operations on the workQueues array can
1194	>	* be checked by comparing runState before vs after the reads.
1195	>	*/
1196	>
1197	>	// bit positions/shifts for fields
1198	>	private static final int AC_SHIFT = 48;
1199	>	private static final int TC_SHIFT = 32;
1200	>	private static final int ST_SHIFT = 31;
1201	>	private static final int EC_SHIFT = 16;
1202	>
1203	>	// bounds
1204	>	private static final int SMASK = 0xffff; // short bits
1205	>	private static final int MAX_CAP = 0x7fff; // max #workers - 1
1206	>	private static final int SQMASK = 0xfffe; // even short bits
1207	>	private static final int SHORT_SIGN = 1 << 15;
1208	>	private static final int INT_SIGN = 1 << 31;
1209	>
1210	>	// masks
1211	>	private static final long STOP_BIT = 0x0001L << ST_SHIFT;
1212	>	private static final long AC_MASK = ((long)SMASK) << AC_SHIFT;
1213	>	private static final long TC_MASK = ((long)SMASK) << TC_SHIFT;
1214	>
1215	>	// units for incrementing and decrementing
1216	>	private static final long TC_UNIT = 1L << TC_SHIFT;
1217	>	private static final long AC_UNIT = 1L << AC_SHIFT;
1218	>
1219	>	// masks and units for dealing with u = (int)(ctl >>> 32)
1220	>	private static final int UAC_SHIFT = AC_SHIFT - 32;
1221	>	private static final int UTC_SHIFT = TC_SHIFT - 32;
1222	>	private static final int UAC_MASK = SMASK << UAC_SHIFT;
1223	>	private static final int UTC_MASK = SMASK << UTC_SHIFT;
1224	>	private static final int UAC_UNIT = 1 << UAC_SHIFT;
1225	>	private static final int UTC_UNIT = 1 << UTC_SHIFT;
1226	>
1227	>	// masks and units for dealing with e = (int)ctl
1228	>	private static final int E_MASK = 0x7fffffff; // no STOP_BIT
1229	>	private static final int E_SEQ = 1 << EC_SHIFT;
1230	>
1231	>	// runState bits
1232	>	private static final int SHUTDOWN = 1 << 31;
1233	>
1234	>	// access mode for WorkQueue
1235	>	static final int LIFO_QUEUE = 0;
1236	>	static final int FIFO_QUEUE = 1;
1237	>	static final int SHARED_QUEUE = -1;
1238	>
1239	>	// Instance fields
1240	>
1241	>	/*
1242	>	* Field layout order in this class tends to matter more than one
1243	>	* would like. Runtime layout order is only loosely related to
1244	>	* declaration order and may differ across JVMs, but the following
1245	>	* empirically works OK on current JVMs.
1246	>	*/
1247	>
1248	>	volatile long ctl; // main pool control
1249	>	final int parallelism; // parallelism level
1250	>	final int localMode; // per-worker scheduling mode
1251	>	final int submitMask; // submit queue index bound
1252	>	int nextSeed; // for initializing worker seeds
1253	>	volatile int runState; // shutdown status and seq
1254	>	WorkQueue[] workQueues; // main registry
1255	>	final Mutex lock; // for registration
1256	>	final Condition termination; // for awaitTermination
1257	>	final ForkJoinWorkerThreadFactory factory; // factory for new workers
1258	>	final Thread.UncaughtExceptionHandler ueh; // per-worker UEH
1259	>	final AtomicLong stealCount; // collect counts when terminated
1260	>	final AtomicInteger nextWorkerNumber; // to create worker name string
1261	>	final String workerNamePrefix; // to create worker name string
1262	>
1263	>	// Creating, registering, and deregistering workers
1264
1265		/**
1266		* Tries to create and start a worker
1267		*/
1268		private void addWorker() {
1269		Throwable ex = null;
1270	<	ForkJoinWorkerThread w = null;
1270	>	ForkJoinWorkerThread wt = null;
1271		try {
1272	<	if ((w = factory.newThread(this)) != null) {
1273	<	w.start();
1272	>	if ((wt = factory.newThread(this)) != null) {
1273	>	wt.start();
1274		return;
1275		}
1276		} catch (Throwable e) {
1277		ex = e;
1278		}
1279	<	deregisterWorker(w, ex);
1279	>	deregisterWorker(wt, ex); // adjust counts etc on failure
1280		}
1281
1282		/**
#	Line 1174 \| Line 1291 \| public class ForkJoinPool extends Abstra
1291		}
1292
1293		/**
1294	<	* Callback from ForkJoinWorkerThread constructor to establish and
1295	<	* record its WorkQueue.
1294	>	* Callback from ForkJoinWorkerThread constructor to establish its
1295	>	* poolIndex and record its WorkQueue. To avoid scanning bias due
1296	>	* to packing entries in front of the workQueues array, we treat
1297	>	* the array as a simple power-of-two hash table using per-thread
1298	>	* seed as hash, expanding as needed.
1299		*
1300	<	* @param wt the worker thread
1300	>	* @param w the worker's queue
1301		*/
1302	<	final void registerWorker(ForkJoinWorkerThread wt) {
1303	<	WorkQueue w = wt.workQueue;
1304	<	ReentrantLock lock = this.lock;
1302	>
1303	>	final void registerWorker(WorkQueue w) {
1304	>	Mutex lock = this.lock;
1305		lock.lock();
1306		try {
1187	–	int k = nextPoolIndex;
1307		WorkQueue[] ws = workQueues;
1308	<	if (ws != null) { // ignore on shutdown
1309	<	int n = ws.length;
1310	<	if (k < 0 \|\| (k & 1) == 0 \|\| k >= n \|\| ws[k] != null) {
1311	<	for (k = 1; k < n && ws[k] != null; k += 2)
1312	<	; // workers are at odd indices
1313	<	if (k >= n) // resize
1314	<	workQueues = ws = Arrays.copyOf(ws, n << 1);
1315	<	}
1316	<	w.poolIndex = k;
1317	<	w.eventCount = ~(k >>> 1) & SMASK; // Set up wait count
1318	<	ws[k] = w; // record worker
1319	<	nextPoolIndex = k + 2;
1320	<	int rs = runState;
1321	<	int m = rs & SMASK; // recalculate runState mask
1322	<	if (k > m)
1323	<	m = (m << 1) + 1;
1324	<	runState = (rs & SHUTDOWN) \| ((rs + RS_SEQ) & RS_SEQ_MASK) \| m;
1308	>	if (w != null && ws != null) { // skip on shutdown/failure
1309	>	int rs, n = ws.length, m = n - 1;
1310	>	int s = nextSeed += SEED_INCREMENT; // rarely-colliding sequence
1311	>	w.seed = (s == 0) ? 1 : s; // ensure non-zero seed
1312	>	int r = (s << 1) \| 1; // use odd-numbered indices
1313	>	if (ws[r &= m] != null) { // collision
1314	>	int probes = 0; // step by approx half size
1315	>	int step = (n <= 4) ? 2 : ((n >>> 1) & SQMASK) + 2;
1316	>	while (ws[r = (r + step) & m] != null) {
1317	>	if (++probes >= n) {
1318	>	workQueues = ws = Arrays.copyOf(ws, n <<= 1);
1319	>	m = n - 1;
1320	>	probes = 0;
1321	>	}
1322	>	}
1323	>	}
1324	>	w.eventCount = w.poolIndex = r; // establish before recording
1325	>	ws[r] = w; // also update seq
1326	>	runState = ((rs = runState) & SHUTDOWN) \| ((rs + 2) & ~SHUTDOWN);
1327		}
1328		} finally {
1329		lock.unlock();
#	Line 1210 \| Line 1331 \| public class ForkJoinPool extends Abstra
1331		}
1332
1333		/**
1334	<	* Final callback from terminating worker, as well as failure to
1335	<	* construct or start a worker in addWorker. Removes record of
1334	>	* Final callback from terminating worker, as well as upon failure
1335	>	* to construct or start a worker in addWorker. Removes record of
1336		* worker from array, and adjusts counts. If pool is shutting
1337		* down, tries to complete termination.
1338		*
#	Line 1219 \| Line 1340 \| public class ForkJoinPool extends Abstra
1340		* @param ex the exception causing failure, or null if none
1341		*/
1342		final void deregisterWorker(ForkJoinWorkerThread wt, Throwable ex) {
1343	+	Mutex lock = this.lock;
1344		WorkQueue w = null;
1345		if (wt != null && (w = wt.workQueue) != null) {
1346		w.runState = -1; // ensure runState is set
1347		stealCount.getAndAdd(w.totalSteals + w.nsteals);
1348		int idx = w.poolIndex;
1227	–	ReentrantLock lock = this.lock;
1349		lock.lock();
1350		try { // remove record from array
1351		WorkQueue[] ws = workQueues;
1352		if (ws != null && idx >= 0 && idx < ws.length && ws[idx] == w)
1353	<	ws[nextPoolIndex = idx] = null;
1353	>	ws[idx] = null;
1354		} finally {
1355		lock.unlock();
1356		}
#	Line 1241 \| Line 1362 \| public class ForkJoinPool extends Abstra
1362		((c - TC_UNIT) & TC_MASK) \|
1363		(c & ~(AC_MASK\|TC_MASK)))));
1364
1365	<	if (!tryTerminate(false) && w != null) {
1365	>	if (!tryTerminate(false, false) && w != null) {
1366		w.cancelAll(); // cancel remaining tasks
1367		if (w.array != null) // suppress signal if never ran
1368		signalWork(); // wake up or create replacement
1369	+	if (ex == null) // help clean refs on way out
1370	+	ForkJoinTask.helpExpungeStaleExceptions();
1371		}
1372
1373		if (ex != null) // rethrow
1374		U.throwException(ex);
1375		}
1376
1377	+
1378	+	// Submissions
1379	+
1380		/**
1381	<	* Tries to add and register a new queue at the given index.
1381	>	* Unless shutting down, adds the given task to a submission queue
1382	>	* at submitter's current queue index (modulo submission
1383	>	* range). If no queue exists at the index, one is created. If
1384	>	* the queue is busy, another index is randomly chosen. The
1385	>	* submitMask bounds the effective number of queues to the
1386	>	* (nearest power of two for) parallelism level.
1387		*
1388	<	* @param idx the workQueues array index to register the queue
1389	<	* @return the queue, or null if could not add because could
1390	<	* not acquire lock or idx is unusable
1391	<	*/
1392	<	private WorkQueue tryAddSharedQueue(int idx) {
1393	<	WorkQueue q = null;
1394	<	ReentrantLock lock = this.lock;
1395	<	if (idx >= 0 && (idx & 1) == 0 && !lock.isLocked()) {
1396	<	// create queue outside of lock but only if apparently free
1397	<	WorkQueue nq = new WorkQueue(null, SHARED_QUEUE);
1398	<	if (lock.tryLock()) {
1399	<	try {
1400	<	WorkQueue[] ws = workQueues;
1401	<	if (ws != null && idx < ws.length) {
1402	<	if ((q = ws[idx]) == null) {
1403	<	int rs; // update runState seq
1404	<	ws[idx] = q = nq;
1405	<	runState = (((rs = runState) & SHUTDOWN) \|
1275	<	((rs + RS_SEQ) & ~SHUTDOWN));
1276	<	}
1388	>	* @param task the task. Caller must ensure non-null.
1389	>	*/
1390	>	private void doSubmit(ForkJoinTask<?> task) {
1391	>	Submitter s = submitters.get();
1392	>	for (int r = s.seed, m = submitMask;;) {
1393	>	WorkQueue[] ws; WorkQueue q;
1394	>	int k = r & m & SQMASK; // use only even indices
1395	>	if (runState < 0 \|\| (ws = workQueues) == null \|\| ws.length <= k)
1396	>	throw new RejectedExecutionException(); // shutting down
1397	>	else if ((q = ws[k]) == null) { // create new queue
1398	>	WorkQueue nq = new WorkQueue(this, null, SHARED_QUEUE);
1399	>	Mutex lock = this.lock; // construct outside lock
1400	>	lock.lock();
1401	>	try { // recheck under lock
1402	>	int rs = runState; // to update seq
1403	>	if (ws == workQueues && ws[k] == null) {
1404	>	ws[k] = nq;
1405	>	runState = ((rs & SHUTDOWN) \| ((rs + 2) & ~SHUTDOWN));
1406		}
1407		} finally {
1408		lock.unlock();
1409		}
1410		}
1411	+	else if (q.trySharedPush(task)) {
1412	+	signalWork();
1413	+	return;
1414	+	}
1415	+	else if (m > 1) { // move to a different index
1416	+	r ^= r << 13; // same xorshift as WorkQueues
1417	+	r ^= r >>> 17;
1418	+	s.seed = r ^= r << 5;
1419	+	}
1420	+	else
1421	+	Thread.yield(); // yield if no alternatives
1422		}
1283	–	return q;
1423		}
1424
1425		// Maintaining ctl counts
#	Line 1294 \| Line 1433 \| public class ForkJoinPool extends Abstra
1433		}
1434
1435		/**
1436	<	* Activates or creates a worker.
1436	>	* Tries to activate or create a worker if too few are active.
1437		*/
1438		final void signalWork() {
1439	<	/*
1440	<	* The while condition is true if: (there is are too few total
1441	<	* workers OR there is at least one waiter) AND (there are too
1442	<	* few active workers OR the pool is terminating). The value
1443	<	* of e distinguishes the remaining cases: zero (no waiters)
1444	<	* for create, negative if terminating (in which case do
1445	<	* nothing), else release a waiter. The secondary checks for
1446	<	* release (non-null array etc) can fail if the pool begins
1447	<	* terminating after the test, and don't impose any added cost
1448	<	* because JVMs must perform null and bounds checks anyway.
1449	<	*/
1450	<	long c; int e, u;
1451	<	while ((((e = (int)(c = ctl)) \| (u = (int)(c >>> 32))) &
1452	<	(INT_SIGN\|SHORT_SIGN)) == (INT_SIGN\|SHORT_SIGN)) {
1314	<	WorkQueue[] ws = workQueues; int i; WorkQueue w; Thread p;
1315	<	if (e == 0) { // add a new worker
1316	<	if (U.compareAndSwapLong
1317	<	(this, CTL, c, (long)(((u + UTC_UNIT) & UTC_MASK) \|
1318	<	((u + UAC_UNIT) & UAC_MASK)) << 32)) {
1319	<	addWorker();
1320	<	break;
1439	>	long c; int u;
1440	>	while ((u = (int)((c = ctl) >>> 32)) < 0) { // too few active
1441	>	WorkQueue[] ws = workQueues; int e, i; WorkQueue w; Thread p;
1442	>	if ((e = (int)c) > 0) { // at least one waiting
1443	>	if (ws != null && (i = e & SMASK) < ws.length &&
1444	>	(w = ws[i]) != null && w.eventCount == (e \| INT_SIGN)) {
1445	>	long nc = (((long)(w.nextWait & E_MASK)) \|
1446	>	((long)(u + UAC_UNIT) << 32));
1447	>	if (U.compareAndSwapLong(this, CTL, c, nc)) {
1448	>	w.eventCount = (e + E_SEQ) & E_MASK;
1449	>	if ((p = w.parker) != null)
1450	>	U.unpark(p); // activate and release
1451	>	break;
1452	>	}
1453		}
1454	<	}
1323	<	else if (e > 0 && ws != null &&
1324	<	(i = ((~e << 1) \| 1) & SMASK) < ws.length &&
1325	<	(w = ws[i]) != null &&
1326	<	w.eventCount == (e \| INT_SIGN)) {
1327	<	if (U.compareAndSwapLong
1328	<	(this, CTL, c, (((long)(w.nextWait & E_MASK)) \|
1329	<	((long)(u + UAC_UNIT) << 32)))) {
1330	<	w.eventCount = (e + E_SEQ) & E_MASK;
1331	<	if ((p = w.parker) != null)
1332	<	U.unpark(p); // release a waiting worker
1454	>	else
1455		break;
1334	–	}
1456		}
1457	<	else
1458	<	break;
1459	<	}
1339	<	}
1340	<
1341	<	/**
1342	<	* Tries to decrement active count (sometimes implicitly) and
1343	<	* possibly release or create a compensating worker in preparation
1344	<	* for blocking. Fails on contention or termination.
1345	<	*
1346	<	* @return true if the caller can block, else should recheck and retry
1347	<	*/
1348	<	final boolean tryCompensate() {
1349	<	WorkQueue[] ws; WorkQueue w; Thread p;
1350	<	int pc = parallelism, e, u, ac, tc, i;
1351	<	long c = ctl;
1352	<
1353	<	if ((e = (int)c) >= 0) {
1354	<	if ((ac = ((u = (int)(c >>> 32)) >> UAC_SHIFT)) <= 0 &&
1355	<	e != 0 && (ws = workQueues) != null &&
1356	<	(i = ((~e << 1) \| 1) & SMASK) < ws.length &&
1357	<	(w = ws[i]) != null) {
1358	<	if (w.eventCount == (e \| INT_SIGN) &&
1359	<	U.compareAndSwapLong
1360	<	(this, CTL, c, ((long)(w.nextWait & E_MASK) \|
1361	<	(c & (AC_MASK\|TC_MASK))))) {
1362	<	w.eventCount = (e + E_SEQ) & E_MASK;
1363	<	if ((p = w.parker) != null)
1364	<	U.unpark(p);
1365	<	return true; // release an idle worker
1366	<	}
1367	<	}
1368	<	else if ((tc = (short)(u >>> UTC_SHIFT)) >= 0 && ac + pc > 1) {
1369	<	long nc = ((c - AC_UNIT) & AC_MASK) \| (c & ~AC_MASK);
1370	<	if (U.compareAndSwapLong(this, CTL, c, nc))
1371	<	return true; // no compensation needed
1372	<	}
1373	<	else if (tc + pc < MAX_ID) {
1374	<	long nc = ((c + TC_UNIT) & TC_MASK) \| (c & ~TC_MASK);
1457	>	else if (e == 0 && (u & SHORT_SIGN) != 0) { // too few total
1458	>	long nc = (long)(((u + UTC_UNIT) & UTC_MASK) \|
1459	>	((u + UAC_UNIT) & UAC_MASK)) << 32;
1460		if (U.compareAndSwapLong(this, CTL, c, nc)) {
1461		addWorker();
1462	<	return true; // create replacement
1462	>	break;
1463		}
1464		}
1465	+	else
1466	+	break;
1467		}
1381	–	return false;
1468		}
1469
1470	<	// Submissions
1470	>	// Scanning for tasks
1471
1472		/**
1473	<	* Unless shutting down, adds the given task to a submission queue
1388	<	* at submitter's current queue index. If no queue exists at the
1389	<	* index, one is created unless pool lock is busy. If the queue
1390	<	* and/or lock are busy, another index is randomly chosen.
1473	>	* Top-level runloop for workers, called by ForkJoinWorkerThread.run.
1474		*/
1475	<	private void doSubmit(ForkJoinTask<?> task) {
1476	<	if (task == null)
1477	<	throw new NullPointerException();
1395	<	Submitter s = submitters.get();
1396	<	for (int r = s.seed;;) {
1397	<	WorkQueue q; int k;
1398	<	int rs = runState, m = rs & SMASK;
1399	<	WorkQueue[] ws = workQueues;
1400	<	if (rs < 0 \|\| ws == null) // shutting down
1401	<	throw new RejectedExecutionException();
1402	<	if (ws.length > m && // k must be at index
1403	<	((q = ws[k = (r << 1) & m]) != null \|\|
1404	<	(q = tryAddSharedQueue(k)) != null) &&
1405	<	q.trySharedPush(task)) {
1406	<	signalWork();
1407	<	return;
1408	<	}
1409	<	r ^= r << 13; // xorshift seed to new position
1410	<	r ^= r >>> 17;
1411	<	if (((s.seed = r ^= r << 5) & m) == 0)
1412	<	Thread.yield(); // occasionally yield if busy
1413	<	}
1475	>	final void runWorker(WorkQueue w) {
1476	>	w.growArray(false); // initialize queue array in this thread
1477	>	do { w.runTask(scan(w)); } while (w.runState >= 0);
1478		}
1479
1416	–
1417	–	// Scanning for tasks
1418	–
1480		/**
1481		* Scans for and, if found, returns one task, else possibly
1482		* inactivates the worker. This method operates on single reads of
1483	<	* volatile state and is designed to be re-invoked continuously in
1484	<	* part because it returns upon detecting inconsistencies,
1483	>	* volatile state and is designed to be re-invoked continuously,
1484	>	* in part because it returns upon detecting inconsistencies,
1485		* contention, or state changes that indicate possible success on
1486		* re-invocation.
1487		*
1488	<	* The scan searches for tasks across queues, randomly selecting
1489	<	* the first #queues probes, favoring steals 2:1 over submissions
1490	<	* (by exploiting even/odd indexing), and then performing a
1491	<	* circular sweep of all queues. The scan terminates upon either
1492	<	* finding a non-empty queue, or completing a full sweep. If the
1493	<	* worker is not inactivated, it takes and returns a task from
1494	<	* this queue. On failure to find a task, we take one of the
1495	<	* following actions, after which the caller will retry calling
1496	<	* this method unless terminated.
1488	>	* The scan searches for tasks across a random permutation of
1489	>	* queues (starting at a random index and stepping by a random
1490	>	* relative prime, checking each at least once). The scan
1491	>	* terminates upon either finding a non-empty queue, or completing
1492	>	* the sweep. If the worker is not inactivated, it takes and
1493	>	* returns a task from this queue. On failure to find a task, we
1494	>	* take one of the following actions, after which the caller will
1495	>	* retry calling this method unless terminated.
1496	>	*
1497	>	* * If pool is terminating, terminate the worker.
1498		*
1499		* * If not a complete sweep, try to release a waiting worker. If
1500		* the scan terminated because the worker is inactivated, then the
#	Line 1441 \| Line 1503 \| public class ForkJoinPool extends Abstra
1503		* another worker, but with same net effect. Releasing in other
1504		* cases as well ensures that we have enough workers running.
1505		*
1444	–	* * If the caller has run a task since the the last empty scan,
1445	–	* return (to allow rescan) if other workers are not also yet
1446	–	* enqueued. Field WorkQueue.rescans counts down on each scan to
1447	–	* ensure eventual inactivation, and occasional calls to
1448	–	* Thread.yield to help avoid interference with more useful
1449	–	* activities on the system.
1450	–	*
1451	–	* * If pool is terminating, terminate the worker.
1452	–	*
1506		* * If not already enqueued, try to inactivate and enqueue the
1507	<	* worker on wait queue.
1507	>	* worker on wait queue. Or, if inactivating has caused the pool
1508	>	* to be quiescent, relay to idleAwaitWork to check for
1509	>	* termination and possibly shrink pool.
1510	>	*
1511	>	* * If already inactive, and the caller has run a task since the
1512	>	* last empty scan, return (to allow rescan) unless others are
1513	>	* also inactivated. Field WorkQueue.rescans counts down on each
1514	>	* scan to ensure eventual inactivation and blocking.
1515		*
1516	<	* * If already enqueued and none of the above apply, either park
1517	<	* awaiting signal, or if this is the most recent waiter and pool
1458	<	* is quiescent, relay to idleAwaitWork to check for termination
1459	<	* and possibly shrink pool.
1516	>	* * If already enqueued and none of the above apply, park
1517	>	* awaiting signal,
1518		*
1519		* @param w the worker (via its WorkQueue)
1520		* @return a task or null of none found
1521		*/
1522		private final ForkJoinTask<?> scan(WorkQueue w) {
1523	<	boolean swept = false; // true after full empty scan
1524	<	WorkQueue[] ws; // volatile read order matters
1525	<	int r = w.seed, ec = w.eventCount; // ec is negative if inactive
1526	<	int rs = runState, m = rs & SMASK;
1527	<	if ((ws = workQueues) != null && ws.length > m) {
1528	<	ForkJoinTask<?> task = null;
1529	<	for (int k = 0, j = -2 - m; ; ++j) {
1530	<	WorkQueue q; int b;
1531	<	if (j < 0) { // random probes while j negative
1532	<	r ^= r << 13; r ^= r >>> 17; k = (r ^= r << 5) \| (j & 1);
1533	<	} // worker (not submit) for odd j
1534	<	else // cyclic scan when j >= 0
1535	<	k += (m >>> 1) \| 1; // step by half to reduce bias
1536	<
1537	<	if ((q = ws[k & m]) != null && (b = q.base) - q.top < 0) {
1538	<	if (ec >= 0)
1539	<	task = q.pollAt(b); // steal
1540	<	break;
1523	>	WorkQueue[] ws; // first update random seed
1524	>	int r = w.seed; r ^= r << 13; r ^= r >>> 17; w.seed = r ^= r << 5;
1525	>	int rs = runState, m; // volatile read order matters
1526	>	if ((ws = workQueues) != null && (m = ws.length - 1) > 0) {
1527	>	int ec = w.eventCount; // ec is negative if inactive
1528	>	int step = (r >>> 16) \| 1; // relative prime
1529	>	for (int j = (m + 1) << 2; ; r += step) {
1530	>	WorkQueue q; ForkJoinTask<?> t; ForkJoinTask<?>[] a; int b;
1531	>	if ((q = ws[r & m]) != null && (b = q.base) - q.top < 0 &&
1532	>	(a = q.array) != null) { // probably nonempty
1533	>	int i = (((a.length - 1) & b) << ASHIFT) + ABASE;
1534	>	t = (ForkJoinTask<?>)U.getObjectVolatile(a, i);
1535	>	if (q.base == b && ec >= 0 && t != null &&
1536	>	U.compareAndSwapObject(a, i, t, null)) {
1537	>	q.base = b + 1; // specialization of pollAt
1538	>	return t;
1539	>	}
1540	>	else if (ec < 0 \|\| j <= m) {
1541	>	rs = 0; // mark scan as imcomplete
1542	>	break; // caller can retry after release
1543	>	}
1544		}
1545	<	else if (j > m) {
1485	<	if (rs == runState) // staleness check
1486	<	swept = true;
1545	>	if (--j < 0)
1546		break;
1547	+	}
1548	+	long c = ctl; int e = (int)c, a = (int)(c >> AC_SHIFT), nr, ns;
1549	+	if (e < 0) // decode ctl on empty scan
1550	+	w.runState = -1; // pool is terminating
1551	+	else if (rs == 0 \|\| rs != runState) { // incomplete scan
1552	+	WorkQueue v; Thread p; // try to release a waiter
1553	+	if (e > 0 && a < 0 && w.eventCount == ec &&
1554	+	(v = ws[e & m]) != null && v.eventCount == (e \| INT_SIGN)) {
1555	+	long nc = ((long)(v.nextWait & E_MASK) \|
1556	+	((c + AC_UNIT) & (AC_MASK\|TC_MASK)));
1557	+	if (ctl == c && U.compareAndSwapLong(this, CTL, c, nc)) {
1558	+	v.eventCount = (e + E_SEQ) & E_MASK;
1559	+	if ((p = v.parker) != null)
1560	+	U.unpark(p);
1561	+	}
1562		}
1563		}
1564	<	w.seed = r; // save seed for next scan
1565	<	if (task != null)
1566	<	return task;
1567	<	}
1568	<
1569	<	// Decode ctl on empty scan
1570	<	long c = ctl; int e = (int)c, a = (int)(c >> AC_SHIFT), nr, ns;
1571	<	if (!swept) { // try to release a waiter
1572	<	WorkQueue v; Thread p;
1573	<	if (e > 0 && a < 0 && ws != null &&
1574	<	(v = ws[((~e << 1) \| 1) & m]) != null &&
1575	<	v.eventCount == (e \| INT_SIGN) && U.compareAndSwapLong
1576	<	(this, CTL, c, ((long)(v.nextWait & E_MASK) \|
1577	<	((c + AC_UNIT) & (AC_MASK\|TC_MASK))))) {
1578	<	v.eventCount = (e + E_SEQ) & E_MASK;
1579	<	if ((p = v.parker) != null)
1580	<	U.unpark(p);
1581	<	}
1582	<	}
1583	<	else if ((nr = w.rescans) > 0) { // continue rescanning
1584	<	int ac = a + parallelism;
1585	<	if ((w.rescans = (ac < nr) ? ac : nr - 1) > 0 && w.seed < 0 &&
1586	<	w.eventCount == ec)
1587	<	Thread.yield(); // 1 bit randomness for yield call
1588	<	}
1589	<	else if (e < 0) // pool is terminating
1590	<	w.runState = -1;
1591	<	else if (ec >= 0) { // try to enqueue
1592	<	long nc = (long)ec \| ((c - AC_UNIT) & (AC_MASK\|TC_MASK));
1593	<	w.nextWait = e;
1594	<	w.eventCount = ec \| INT_SIGN; // mark as inactive
1595	<	if (!U.compareAndSwapLong(this, CTL, c, nc))
1522	<	w.eventCount = ec; // back out on CAS failure
1523	<	else if ((ns = w.nsteals) != 0) { // set rescans if ran task
1524	<	if (a <= 0) // ... unless too many active
1525	<	w.rescans = a + parallelism;
1526	<	w.nsteals = 0;
1527	<	w.totalSteals += ns;
1528	<	}
1529	<	}
1530	<	else{ // already queued
1531	<	if (parallelism == -a)
1532	<	idleAwaitWork(w); // quiescent
1533	<	if (w.eventCount == ec) {
1534	<	Thread.interrupted(); // clear status
1535	<	ForkJoinWorkerThread wt = w.owner;
1536	<	U.putObject(wt, PARKBLOCKER, this);
1537	<	w.parker = wt; // emulate LockSupport.park
1538	<	if (w.eventCount == ec) // recheck
1539	<	U.park(false, 0L); // block
1540	<	w.parker = null;
1541	<	U.putObject(wt, PARKBLOCKER, null);
1564	>	else if (ec >= 0) { // try to enqueue/inactivate
1565	>	long nc = (long)ec \| ((c - AC_UNIT) & (AC_MASK\|TC_MASK));
1566	>	w.nextWait = e;
1567	>	w.eventCount = ec \| INT_SIGN; // mark as inactive
1568	>	if (ctl != c \|\| !U.compareAndSwapLong(this, CTL, c, nc))
1569	>	w.eventCount = ec; // unmark on CAS failure
1570	>	else {
1571	>	if ((ns = w.nsteals) != 0) {
1572	>	w.nsteals = 0; // set rescans if ran task
1573	>	w.rescans = (a > 0) ? 0 : a + parallelism;
1574	>	w.totalSteals += ns;
1575	>	}
1576	>	if (a == 1 - parallelism) // quiescent
1577	>	idleAwaitWork(w, nc, c);
1578	>	}
1579	>	}
1580	>	else if (w.eventCount < 0) { // already queued
1581	>	if ((nr = w.rescans) > 0) { // continue rescanning
1582	>	int ac = a + parallelism;
1583	>	if (((w.rescans = (ac < nr) ? ac : nr - 1) & 3) == 0)
1584	>	Thread.yield(); // yield before block
1585	>	}
1586	>	else {
1587	>	Thread.interrupted(); // clear status
1588	>	Thread wt = Thread.currentThread();
1589	>	U.putObject(wt, PARKBLOCKER, this);
1590	>	w.parker = wt; // emulate LockSupport.park
1591	>	if (w.eventCount < 0) // recheck
1592	>	U.park(false, 0L);
1593	>	w.parker = null;
1594	>	U.putObject(wt, PARKBLOCKER, null);
1595	>	}
1596		}
1597		}
1598		return null;
1599		}
1600
1601		/**
1602	<	* If inactivating worker w has caused pool to become quiescent,
1603	<	* checks for pool termination, and, so long as this is not the
1604	<	* only worker, waits for event for up to SHRINK_RATE nanosecs.
1605	<	* On timeout, if ctl has not changed, terminates the worker,
1606	<	* which will in turn wake up another worker to possibly repeat
1607	<	* this process.
1602	>	* If inactivating worker w has caused the pool to become
1603	>	* quiescent, checks for pool termination, and, so long as this is
1604	>	* not the only worker, waits for event for up to SHRINK_RATE
1605	>	* nanosecs. On timeout, if ctl has not changed, terminates the
1606	>	* worker, which will in turn wake up another worker to possibly
1607	>	* repeat this process.
1608		*
1609		* @param w the calling worker
1610	+	* @param currentCtl the ctl value triggering possible quiescence
1611	+	* @param prevCtl the ctl value to restore if thread is terminated
1612		*/
1613	<	private void idleAwaitWork(WorkQueue w) {
1614	<	long c; int nw, ec;
1615	<	if (!tryTerminate(false) &&
1616	<	(int)((c = ctl) >> AC_SHIFT) + parallelism == 0 &&
1617	<	(ec = w.eventCount) == ((int)c \| INT_SIGN) &&
1618	<	(nw = w.nextWait) != 0) {
1563	<	long nc = ((long)(nw & E_MASK) \| // ctl to restore on timeout
1564	<	((c + AC_UNIT) & AC_MASK) \| (c & TC_MASK));
1565	<	ForkJoinTask.helpExpungeStaleExceptions(); // help clean
1566	<	ForkJoinWorkerThread wt = w.owner;
1567	<	while (ctl == c) {
1613	>	private void idleAwaitWork(WorkQueue w, long currentCtl, long prevCtl) {
1614	>	if (w.eventCount < 0 && !tryTerminate(false, false) &&
1615	>	(int)prevCtl != 0 && !hasQueuedSubmissions() && ctl == currentCtl) {
1616	>	Thread wt = Thread.currentThread();
1617	>	Thread.yield(); // yield before block
1618	>	while (ctl == currentCtl) {
1619		long startTime = System.nanoTime();
1620		Thread.interrupted(); // timed variant of version in scan()
1621		U.putObject(wt, PARKBLOCKER, this);
1622		w.parker = wt;
1623	<	if (ctl == c)
1623	>	if (ctl == currentCtl)
1624		U.park(false, SHRINK_RATE);
1625		w.parker = null;
1626		U.putObject(wt, PARKBLOCKER, null);
1627	<	if (ctl != c)
1627	>	if (ctl != currentCtl)
1628		break;
1629		if (System.nanoTime() - startTime >= SHRINK_TIMEOUT &&
1630	<	U.compareAndSwapLong(this, CTL, c, nc)) {
1631	<	w.runState = -1; // shrink
1632	<	w.eventCount = (ec + E_SEQ) \| E_MASK;
1630	>	U.compareAndSwapLong(this, CTL, currentCtl, prevCtl)) {
1631	>	w.eventCount = (w.eventCount + E_SEQ) \| E_MASK;
1632	>	w.runState = -1; // shrink
1633		break;
1634		}
1635		}
#	Line 1596 \| Line 1647 \| public class ForkJoinPool extends Abstra
1647		* leaves hints in workers to speed up subsequent calls. The
1648		* implementation is very branchy to cope with potential
1649		* inconsistencies or loops encountering chains that are stale,
1650	<	* unknown, or of length greater than MAX_HELP_DEPTH links. All
1651	<	* of these cases are dealt with by just retrying by caller.
1650	>	* unknown, or so long that they are likely cyclic. All of these
1651	>	* cases are dealt with by just retrying by caller.
1652		*
1653		* @param joiner the joining worker
1654		* @param task the task to join
1655		* @return true if found or ran a task (and so is immediately retryable)
1656		*/
1657	<	final boolean tryHelpStealer(WorkQueue joiner, ForkJoinTask<?> task) {
1658	<	ForkJoinTask<?> subtask; // current target
1657	>	private boolean tryHelpStealer(WorkQueue joiner, ForkJoinTask<?> task) {
1658	>	WorkQueue[] ws;
1659	>	int m, depth = MAX_HELP; // remaining chain depth
1660		boolean progress = false;
1661	<	int depth = 0; // current chain depth
1662	<	int m = runState & SMASK;
1663	<	WorkQueue[] ws = workQueues;
1664	<
1665	<	if (ws != null && ws.length > m && (subtask = task).status >= 0) {
1666	<	outer:for (WorkQueue j = joiner;;) {
1615	<	// Try to find the stealer of subtask, by first using hint
1616	<	WorkQueue stealer = null;
1617	<	WorkQueue v = ws[j.stealHint & m];
1661	>	if ((ws = workQueues) != null && (m = ws.length - 1) > 0 &&
1662	>	task.status >= 0) {
1663	>	ForkJoinTask<?> subtask = task; // current target
1664	>	outer: for (WorkQueue j = joiner;;) {
1665	>	WorkQueue stealer = null; // find stealer of subtask
1666	>	WorkQueue v = ws[j.stealHint & m]; // try hint
1667		if (v != null && v.currentSteal == subtask)
1668		stealer = v;
1669	<	else {
1669	>	else { // scan
1670		for (int i = 1; i <= m; i += 2) {
1671	<	if ((v = ws[i]) != null && v.currentSteal == subtask) {
1671	>	if ((v = ws[i]) != null && v.currentSteal == subtask &&
1672	>	v != joiner) {
1673		stealer = v;
1674	<	j.stealHint = i; // save hint
1674	>	j.stealHint = i; // save hint
1675		break;
1676		}
1677		}
#	Line 1629 \| Line 1679 \| public class ForkJoinPool extends Abstra
1679		break;
1680		}
1681
1682	<	for (WorkQueue q = stealer;;) { // Try to help stealer
1683	<	ForkJoinTask<?> t; int b;
1682	>	for (WorkQueue q = stealer;;) { // try to help stealer
1683	>	ForkJoinTask[] a; ForkJoinTask<?> t; int b;
1684		if (task.status < 0)
1685		break outer;
1686	<	if ((b = q.base) - q.top < 0) {
1686	>	if ((b = q.base) - q.top < 0 && (a = q.array) != null) {
1687		progress = true;
1688	<	if (subtask.status < 0)
1689	<	break outer; // stale
1690	<	if ((t = q.pollAt(b)) != null) {
1691	<	stealer.stealHint = joiner.poolIndex;
1688	>	int i = (((a.length - 1) & b) << ASHIFT) + ABASE;
1689	>	t = (ForkJoinTask<?>)U.getObjectVolatile(a, i);
1690	>	if (subtask.status < 0) // must recheck before taking
1691	>	break outer;
1692	>	if (t != null &&
1693	>	q.base == b &&
1694	>	U.compareAndSwapObject(a, i, t, null)) {
1695	>	q.base = b + 1;
1696		joiner.runSubtask(t);
1697		}
1698	+	else if (q.base == b)
1699	+	break outer; // possibly stalled
1700		}
1701	<	else { // empty - try to descend to find stealer's stealer
1701	>	else { // descend
1702		ForkJoinTask<?> next = stealer.currentJoin;
1703	<	if (++depth == MAX_HELP_DEPTH \|\| subtask.status < 0 \|\|
1703	>	if (--depth <= 0 \|\| subtask.status < 0 \|\|
1704		next == null \|\| next == subtask)
1705	<	break outer; // max depth, stale, dead-end, cyclic
1705	>	break outer; // stale, dead-end, or cyclic
1706		subtask = next;
1707		j = stealer;
1708		break;
#	Line 1663 \| Line 1719 \| public class ForkJoinPool extends Abstra
1719		* @param joiner the joining worker
1720		* @param task the task
1721		*/
1722	<	final void tryPollForAndExec(WorkQueue joiner, ForkJoinTask<?> task) {
1722	>	private void tryPollForAndExec(WorkQueue joiner, ForkJoinTask<?> task) {
1723		WorkQueue[] ws;
1724	<	int m = runState & SMASK;
1725	<	if ((ws = workQueues) != null && ws.length > m) {
1670	<	for (int j = 1; j <= m && task.status >= 0; j += 2) {
1724	>	if ((ws = workQueues) != null) {
1725	>	for (int j = 1; j < ws.length && task.status >= 0; j += 2) {
1726		WorkQueue q = ws[j];
1727		if (q != null && q.pollFor(task)) {
1728		joiner.runSubtask(task);
#	Line 1678 \| Line 1733 \| public class ForkJoinPool extends Abstra
1733		}
1734
1735		/**
1736	<	* Returns a non-empty steal queue, if one is found during a random,
1737	<	* then cyclic scan, else null. This method must be retried by
1738	<	* caller if, by the time it tries to use the queue, it is empty.
1736	>	* Tries to decrement active count (sometimes implicitly) and
1737	>	* possibly release or create a compensating worker in preparation
1738	>	* for blocking. Fails on contention or termination. Otherwise,
1739	>	* adds a new thread if no idle workers are available and either
1740	>	* pool would become completely starved or: (at least half
1741	>	* starved, and fewer than 50% spares exist, and there is at least
1742	>	* one task apparently available). Even though the availability
1743	>	* check requires a full scan, it is worthwhile in reducing false
1744	>	* alarms.
1745	>	*
1746	>	* @param task if non-null, a task being waited for
1747	>	* @param blocker if non-null, a blocker being waited for
1748	>	* @return true if the caller can block, else should recheck and retry
1749	>	*/
1750	>	final boolean tryCompensate(ForkJoinTask<?> task, ManagedBlocker blocker) {
1751	>	int pc = parallelism, e;
1752	>	long c = ctl;
1753	>	WorkQueue[] ws = workQueues;
1754	>	if ((e = (int)c) >= 0 && ws != null) {
1755	>	int u, a, ac, hc;
1756	>	int tc = (short)((u = (int)(c >>> 32)) >>> UTC_SHIFT) + pc;
1757	>	boolean replace = false;
1758	>	if ((a = u >> UAC_SHIFT) <= 0) {
1759	>	if ((ac = a + pc) <= 1)
1760	>	replace = true;
1761	>	else if ((e > 0 \|\| (task != null &&
1762	>	ac <= (hc = pc >>> 1) && tc < pc + hc))) {
1763	>	WorkQueue w;
1764	>	for (int j = 0; j < ws.length; ++j) {
1765	>	if ((w = ws[j]) != null && !w.isEmpty()) {
1766	>	replace = true;
1767	>	break; // in compensation range and tasks available
1768	>	}
1769	>	}
1770	>	}
1771	>	}
1772	>	if ((task == null \|\| task.status >= 0) && // recheck need to block
1773	>	(blocker == null \|\| !blocker.isReleasable()) && ctl == c) {
1774	>	if (!replace) { // no compensation
1775	>	long nc = ((c - AC_UNIT) & AC_MASK) \| (c & ~AC_MASK);
1776	>	if (U.compareAndSwapLong(this, CTL, c, nc))
1777	>	return true;
1778	>	}
1779	>	else if (e != 0) { // release an idle worker
1780	>	WorkQueue w; Thread p; int i;
1781	>	if ((i = e & SMASK) < ws.length && (w = ws[i]) != null) {
1782	>	long nc = ((long)(w.nextWait & E_MASK) \|
1783	>	(c & (AC_MASK\|TC_MASK)));
1784	>	if (w.eventCount == (e \| INT_SIGN) &&
1785	>	U.compareAndSwapLong(this, CTL, c, nc)) {
1786	>	w.eventCount = (e + E_SEQ) & E_MASK;
1787	>	if ((p = w.parker) != null)
1788	>	U.unpark(p);
1789	>	return true;
1790	>	}
1791	>	}
1792	>	}
1793	>	else if (tc < MAX_CAP) { // create replacement
1794	>	long nc = ((c + TC_UNIT) & TC_MASK) \| (c & ~TC_MASK);
1795	>	if (U.compareAndSwapLong(this, CTL, c, nc)) {
1796	>	addWorker();
1797	>	return true;
1798	>	}
1799	>	}
1800	>	}
1801	>	}
1802	>	return false;
1803	>	}
1804	>
1805	>	/**
1806	>	* Helps and/or blocks until the given task is done.
1807	>	*
1808	>	* @param joiner the joining worker
1809	>	* @param task the task
1810	>	* @return task status on exit
1811	>	*/
1812	>	final int awaitJoin(WorkQueue joiner, ForkJoinTask<?> task) {
1813	>	int s;
1814	>	ForkJoinTask<?> prevJoin = joiner.currentJoin;
1815	>	if ((s = task.status) >= 0) {
1816	>	joiner.currentJoin = task;
1817	>	long startTime = 0L;
1818	>	for (int k = 0;;) {
1819	>	if ((joiner.isEmpty() ? // try to help
1820	>	!tryHelpStealer(joiner, task) :
1821	>	!joiner.tryRemoveAndExec(task))) {
1822	>	if (k == 0) {
1823	>	startTime = System.nanoTime();
1824	>	tryPollForAndExec(joiner, task); // check uncommon case
1825	>	}
1826	>	else if ((k & (MAX_HELP - 1)) == 0 &&
1827	>	System.nanoTime() - startTime >=
1828	>	COMPENSATION_DELAY &&
1829	>	tryCompensate(task, null)) {
1830	>	if (task.trySetSignal() && task.status >= 0) {
1831	>	synchronized (task) {
1832	>	if (task.status >= 0) {
1833	>	try { // see ForkJoinTask
1834	>	task.wait(); // for explanation
1835	>	} catch (InterruptedException ie) {
1836	>	}
1837	>	}
1838	>	else
1839	>	task.notifyAll();
1840	>	}
1841	>	}
1842	>	long c; // re-activate
1843	>	do {} while (!U.compareAndSwapLong
1844	>	(this, CTL, c = ctl, c + AC_UNIT));
1845	>	}
1846	>	}
1847	>	if ((s = task.status) < 0) {
1848	>	joiner.currentJoin = prevJoin;
1849	>	break;
1850	>	}
1851	>	else if ((k++ & (MAX_HELP - 1)) == MAX_HELP >>> 1)
1852	>	Thread.yield(); // for politeness
1853	>	}
1854	>	}
1855	>	return s;
1856	>	}
1857	>
1858	>	/**
1859	>	* Stripped-down variant of awaitJoin used by timed joins. Tries
1860	>	* to help join only while there is continuous progress. (Caller
1861	>	* will then enter a timed wait.)
1862	>	*
1863	>	* @param joiner the joining worker
1864	>	* @param task the task
1865	>	* @return task status on exit
1866	>	*/
1867	>	final int helpJoinOnce(WorkQueue joiner, ForkJoinTask<?> task) {
1868	>	int s;
1869	>	while ((s = task.status) >= 0 &&
1870	>	(joiner.isEmpty() ?
1871	>	tryHelpStealer(joiner, task) :
1872	>	joiner.tryRemoveAndExec(task)))
1873	>	;
1874	>	return s;
1875	>	}
1876	>
1877	>	/**
1878	>	* Returns a (probably) non-empty steal queue, if one is found
1879	>	* during a random, then cyclic scan, else null. This method must
1880	>	* be retried by caller if, by the time it tries to use the queue,
1881	>	* it is empty.
1882		*/
1883		private WorkQueue findNonEmptyStealQueue(WorkQueue w) {
1884	<	int r = w.seed; // Same idea as scan(), but ignoring submissions
1884	>	// Similar to loop in scan(), but ignoring submissions
1885	>	int r = w.seed; r ^= r << 13; r ^= r >>> 17; w.seed = r ^= r << 5;
1886	>	int step = (r >>> 16) \| 1;
1887		for (WorkQueue[] ws;;) {
1888	<	int m = runState & SMASK;
1889	<	if ((ws = workQueues) == null)
1888	>	int rs = runState, m;
1889	>	if ((ws = workQueues) == null \|\| (m = ws.length - 1) < 1)
1890		return null;
1891	<	if (ws.length > m) {
1892	<	WorkQueue q;
1893	<	for (int n = m << 2, k = r, j = -n;;) {
1894	<	r ^= r << 13; r ^= r >>> 17; r ^= r << 5;
1895	<	if ((q = ws[(k \| 1) & m]) != null && q.base - q.top < 0) {
1896	<	w.seed = r;
1697	<	return q;
1698	<	}
1699	<	else if (j > n)
1891	>	for (int j = (m + 1) << 2; ; r += step) {
1892	>	WorkQueue q = ws[((r << 1) \| 1) & m];
1893	>	if (q != null && !q.isEmpty())
1894	>	return q;
1895	>	else if (--j < 0) {
1896	>	if (runState == rs)
1897		return null;
1898	<	else
1702	<	k = (j++ < 0) ? r : k + ((m >>> 1) \| 1);
1703	<
1898	>	break;
1899		}
1900		}
1901		}
#	Line 1714 \| Line 1909 \| public class ForkJoinPool extends Abstra
1909		*/
1910		final void helpQuiescePool(WorkQueue w) {
1911		for (boolean active = true;;) {
1912	<	w.runLocalTasks(); // exhaust local queue
1912	>	ForkJoinTask<?> localTask; // exhaust local queue
1913	>	while ((localTask = w.nextLocalTask()) != null)
1914	>	localTask.doExec();
1915		WorkQueue q = findNonEmptyStealQueue(w);
1916		if (q != null) {
1917	<	ForkJoinTask<?> t;
1917	>	ForkJoinTask<?> t; int b;
1918		if (!active) { // re-establish active count
1919		long c;
1920		active = true;
1921		do {} while (!U.compareAndSwapLong
1922		(this, CTL, c = ctl, c + AC_UNIT));
1923		}
1924	<	if ((t = q.poll()) != null)
1924	>	if ((b = q.base) - q.top < 0 && (t = q.pollAt(b)) != null)
1925		w.runSubtask(t);
1926		}
1927		else {
#	Line 1752 \| Line 1949 \| public class ForkJoinPool extends Abstra
1949		*/
1950		final ForkJoinTask<?> nextTaskFor(WorkQueue w) {
1951		for (ForkJoinTask<?> t;;) {
1952	<	WorkQueue q;
1952	>	WorkQueue q; int b;
1953		if ((t = w.nextLocalTask()) != null)
1954		return t;
1955		if ((q = findNonEmptyStealQueue(w)) == null)
1956		return null;
1957	<	if ((t = q.poll()) != null)
1957	>	if ((b = q.base) - q.top < 0 && (t = q.pollAt(b)) != null)
1958		return t;
1959		}
1960		}
#	Line 1778 \| Line 1975 \| public class ForkJoinPool extends Abstra
1975		8);
1976		}
1977
1978	<	// Termination
1978	>	// Termination
1979
1980		/**
1981	<	* Sets SHUTDOWN bit of runState under lock
1982	<	*/
1983	<	private void enableShutdown() {
1984	<	ReentrantLock lock = this.lock;
1985	<	if (runState >= 0) {
1986	<	lock.lock(); // don't need try/finally
1987	<	runState \|= SHUTDOWN;
1791	<	lock.unlock();
1792	<	}
1793	<	}
1794	<
1795	<	/**
1796	<	* Possibly initiates and/or completes termination. Upon
1797	<	* termination, cancels all queued tasks and then
1981	>	* Possibly initiates and/or completes termination. The caller
1982	>	* triggering termination runs three passes through workQueues:
1983	>	* (0) Setting termination status, followed by wakeups of queued
1984	>	* workers; (1) cancelling all tasks; (2) interrupting lagging
1985	>	* threads (likely in external tasks, but possibly also blocked in
1986	>	* joins). Each pass repeats previous steps because of potential
1987	>	* lagging thread creation.
1988		*
1989		* @param now if true, unconditionally terminate, else only
1990		* if no work and no active workers
1991	+	* @param enable if true, enable shutdown when next possible
1992		* @return true if now terminating or terminated
1993		*/
1994	<	private boolean tryTerminate(boolean now) {
1994	>	private boolean tryTerminate(boolean now, boolean enable) {
1995	>	Mutex lock = this.lock;
1996		for (long c;;) {
1997		if (((c = ctl) & STOP_BIT) != 0) { // already terminating
1998		if ((short)(c >>> TC_SHIFT) == -parallelism) {
1807	–	ReentrantLock lock = this.lock; // signal when no workers
1999		lock.lock(); // don't need try/finally
2000		termination.signalAll(); // signal when 0 workers
2001		lock.unlock();
2002		}
2003		return true;
2004		}
2005	<	if (!now) {
2006	<	if ((int)(c >> AC_SHIFT) != -parallelism \|\| runState >= 0 \|\|
2005	>	if (runState >= 0) { // not yet enabled
2006	>	if (!enable)
2007	>	return false;
2008	>	lock.lock();
2009	>	runState \|= SHUTDOWN;
2010	>	lock.unlock();
2011	>	}
2012	>	if (!now) { // check if idle & no tasks
2013	>	if ((int)(c >> AC_SHIFT) != -parallelism \|\|
2014		hasQueuedSubmissions())
2015		return false;
2016		// Check for unqueued inactive workers. One pass suffices.
2017		WorkQueue[] ws = workQueues; WorkQueue w;
2018		if (ws != null) {
2019	<	int n = ws.length;
1822	<	for (int i = 1; i < n; i += 2) {
2019	>	for (int i = 1; i < ws.length; i += 2) {
2020		if ((w = ws[i]) != null && w.eventCount >= 0)
2021		return false;
2022		}
2023		}
2024		}
2025	<	if (U.compareAndSwapLong(this, CTL, c, c \| STOP_BIT))
2026	<	startTerminating();
2027	<	}
2028	<	}
2029	<
2030	<	/**
2031	<	* Initiates termination: Runs three passes through workQueues:
2032	<	* (0) Setting termination status, followed by wakeups of queued
2033	<	* workers; (1) cancelling all tasks; (2) interrupting lagging
2034	<	* threads (likely in external tasks, but possibly also blocked in
2035	<	* joins). Each pass repeats previous steps because of potential
2036	<	* lagging thread creation.
2037	<	*/
1841	<	private void startTerminating() {
1842	<	for (int pass = 0; pass < 3; ++pass) {
1843	<	WorkQueue[] ws = workQueues;
1844	<	if (ws != null) {
1845	<	WorkQueue w; Thread wt;
1846	<	int n = ws.length;
1847	<	for (int j = 0; j < n; ++j) {
1848	<	if ((w = ws[j]) != null) {
1849	<	w.runState = -1;
1850	<	if (pass > 0) {
1851	<	w.cancelAll();
1852	<	if (pass > 1 && (wt = w.owner) != null &&
1853	<	!wt.isInterrupted()) {
1854	<	try {
1855	<	wt.interrupt();
1856	<	} catch (SecurityException ignore) {
2025	>	if (U.compareAndSwapLong(this, CTL, c, c \| STOP_BIT)) {
2026	>	for (int pass = 0; pass < 3; ++pass) {
2027	>	WorkQueue[] ws = workQueues;
2028	>	if (ws != null) {
2029	>	WorkQueue w;
2030	>	int n = ws.length;
2031	>	for (int i = 0; i < n; ++i) {
2032	>	if ((w = ws[i]) != null) {
2033	>	w.runState = -1;
2034	>	if (pass > 0) {
2035	>	w.cancelAll();
2036	>	if (pass > 1)
2037	>	w.interruptOwner();
2038		}
2039		}
2040		}
2041	<	}
2042	<	}
2043	<	// Wake up workers parked on event queue
2044	<	int i, e; long c; Thread p;
2045	<	while ((i = ((~(e = (int)(c = ctl)) << 1) \| 1) & SMASK) < n &&
2046	<	(w = ws[i]) != null &&
2047	<	w.eventCount == (e \| INT_SIGN)) {
2048	<	long nc = ((long)(w.nextWait & E_MASK) \|
2049	<	((c + AC_UNIT) & AC_MASK) \|
2050	<	(c & (TC_MASK\|STOP_BIT)));
2051	<	if (U.compareAndSwapLong(this, CTL, c, nc)) {
2052	<	w.eventCount = (e + E_SEQ) & E_MASK;
2053	<	if ((p = w.parker) != null)
2054	<	U.unpark(p);
2041	>	// Wake up workers parked on event queue
2042	>	int i, e; long cc; Thread p;
2043	>	while ((e = (int)(cc = ctl) & E_MASK) != 0 &&
2044	>	(i = e & SMASK) < n &&
2045	>	(w = ws[i]) != null) {
2046	>	long nc = ((long)(w.nextWait & E_MASK) \|
2047	>	((cc + AC_UNIT) & AC_MASK) \|
2048	>	(cc & (TC_MASK\|STOP_BIT)));
2049	>	if (w.eventCount == (e \| INT_SIGN) &&
2050	>	U.compareAndSwapLong(this, CTL, cc, nc)) {
2051	>	w.eventCount = (e + E_SEQ) & E_MASK;
2052	>	w.runState = -1;
2053	>	if ((p = w.parker) != null)
2054	>	U.unpark(p);
2055	>	}
2056	>	}
2057		}
2058		}
2059		}
#	Line 1946 \| Line 2129 \| public class ForkJoinPool extends Abstra
2129		checkPermission();
2130		if (factory == null)
2131		throw new NullPointerException();
2132	<	if (parallelism <= 0 \|\| parallelism > MAX_ID)
2132	>	if (parallelism <= 0 \|\| parallelism > MAX_CAP)
2133		throw new IllegalArgumentException();
2134		this.parallelism = parallelism;
2135		this.factory = factory;
2136		this.ueh = handler;
2137		this.localMode = asyncMode ? FIFO_QUEUE : LIFO_QUEUE;
1955	–	this.nextPoolIndex = 1;
2138		long np = (long)(-parallelism); // offset ctl counts
2139		this.ctl = ((np << AC_SHIFT) & AC_MASK) \| ((np << TC_SHIFT) & TC_MASK);
2140	<	// initialize workQueues array with room for 2*parallelism if possible
2141	<	int n = parallelism << 1;
2142	<	if (n >= MAX_ID)
2143	<	n = MAX_ID;
2144	<	else { // See Hackers Delight, sec 3.2, where n < (1 << 16)
2145	<	n \|= n >>> 1; n \|= n >>> 2; n \|= n >>> 4; n \|= n >>> 8;
2146	<	}
1965	<	this.workQueues = new WorkQueue[(n + 1) << 1];
1966	<	ReentrantLock lck = this.lock = new ReentrantLock();
1967	<	this.termination = lck.newCondition();
2140	>	// Use nearest power 2 for workQueues size. See Hackers Delight sec 3.2.
2141	>	int n = parallelism - 1;
2142	>	n \|= n >>> 1; n \|= n >>> 2; n \|= n >>> 4; n \|= n >>> 8; n \|= n >>> 16;
2143	>	int size = (n + 1) << 1; // #slots = 2*#workers
2144	>	this.submitMask = size - 1; // room for max # of submit queues
2145	>	this.workQueues = new WorkQueue[size];
2146	>	this.termination = (this.lock = new Mutex()).newCondition();
2147		this.stealCount = new AtomicLong();
2148		this.nextWorkerNumber = new AtomicInteger();
2149	+	int pn = poolNumberGenerator.incrementAndGet();
2150		StringBuilder sb = new StringBuilder("ForkJoinPool-");
2151	<	sb.append(poolNumberGenerator.incrementAndGet());
2151	>	sb.append(Integer.toString(pn));
2152		sb.append("-worker-");
2153		this.workerNamePrefix = sb.toString();
2154	<	// Create initial submission queue
2155	<	WorkQueue sq = tryAddSharedQueue(0);
2156	<	if (sq != null)
1977	<	sq.growArray(false);
2154	>	lock.lock();
2155	>	this.runState = 1; // set init flag
2156	>	lock.unlock();
2157		}
2158
2159		// Execution methods
#	Line 1996 \| Line 2175 \| public class ForkJoinPool extends Abstra
2175		* scheduled for execution
2176		*/
2177		public <T> T invoke(ForkJoinTask<T> task) {
2178	+	if (task == null)
2179	+	throw new NullPointerException();
2180		doSubmit(task);
2181		return task.join();
2182		}
#	Line 2009 \| Line 2190 \| public class ForkJoinPool extends Abstra
2190		* scheduled for execution
2191		*/
2192		public void execute(ForkJoinTask<?> task) {
2193	+	if (task == null)
2194	+	throw new NullPointerException();
2195		doSubmit(task);
2196		}
2197
#	Line 2026 \| Line 2209 \| public class ForkJoinPool extends Abstra
2209		if (task instanceof ForkJoinTask<?>) // avoid re-wrap
2210		job = (ForkJoinTask<?>) task;
2211		else
2212	<	job = ForkJoinTask.adapt(task, null);
2212	>	job = new ForkJoinTask.AdaptedRunnableAction(task);
2213		doSubmit(job);
2214		}
2215
#	Line 2040 \| Line 2223 \| public class ForkJoinPool extends Abstra
2223		* scheduled for execution
2224		*/
2225		public <T> ForkJoinTask<T> submit(ForkJoinTask<T> task) {
2226	+	if (task == null)
2227	+	throw new NullPointerException();
2228		doSubmit(task);
2229		return task;
2230		}
#	Line 2050 \| Line 2235 \| public class ForkJoinPool extends Abstra
2235		* scheduled for execution
2236		*/
2237		public <T> ForkJoinTask<T> submit(Callable<T> task) {
2238	<	if (task == null)
2054	<	throw new NullPointerException();
2055	<	ForkJoinTask<T> job = ForkJoinTask.adapt(task);
2238	>	ForkJoinTask<T> job = new ForkJoinTask.AdaptedCallable<T>(task);
2239		doSubmit(job);
2240		return job;
2241		}
#	Line 2063 \| Line 2246 \| public class ForkJoinPool extends Abstra
2246		* scheduled for execution
2247		*/
2248		public <T> ForkJoinTask<T> submit(Runnable task, T result) {
2249	<	if (task == null)
2067	<	throw new NullPointerException();
2068	<	ForkJoinTask<T> job = ForkJoinTask.adapt(task, result);
2249	>	ForkJoinTask<T> job = new ForkJoinTask.AdaptedRunnable<T>(task, result);
2250		doSubmit(job);
2251		return job;
2252		}
#	Line 2082 \| Line 2263 \| public class ForkJoinPool extends Abstra
2263		if (task instanceof ForkJoinTask<?>) // avoid re-wrap
2264		job = (ForkJoinTask<?>) task;
2265		else
2266	<	job = ForkJoinTask.adapt(task, null);
2266	>	job = new ForkJoinTask.AdaptedRunnableAction(task);
2267		doSubmit(job);
2268		return job;
2269		}
#	Line 2092 \| Line 2273 \| public class ForkJoinPool extends Abstra
2273		* @throws RejectedExecutionException {@inheritDoc}
2274		*/
2275		public <T> List<Future<T>> invokeAll(Collection<? extends Callable<T>> tasks) {
2276	<	ArrayList<ForkJoinTask<T>> forkJoinTasks =
2277	<	new ArrayList<ForkJoinTask<T>>(tasks.size());
2278	<	for (Callable<T> task : tasks)
2279	<	forkJoinTasks.add(ForkJoinTask.adapt(task));
2280	<	invoke(new InvokeAll<T>(forkJoinTasks));
2281	<
2276	>	// In previous versions of this class, this method constructed
2277	>	// a task to run ForkJoinTask.invokeAll, but now external
2278	>	// invocation of multiple tasks is at least as efficient.
2279	>	List<ForkJoinTask<T>> fs = new ArrayList<ForkJoinTask<T>>(tasks.size());
2280	>	// Workaround needed because method wasn't declared with
2281	>	// wildcards in return type but should have been.
2282		@SuppressWarnings({"unchecked", "rawtypes"})
2283	<	List<Future<T>> futures = (List<Future<T>>) (List) forkJoinTasks;
2103	<	return futures;
2104	<	}
2283	>	List<Future<T>> futures = (List<Future<T>>) (List) fs;
2284
2285	<	static final class InvokeAll<T> extends RecursiveAction {
2286	<	final ArrayList<ForkJoinTask<T>> tasks;
2287	<	InvokeAll(ArrayList<ForkJoinTask<T>> tasks) { this.tasks = tasks; }
2288	<	public void compute() {
2289	<	try { invokeAll(tasks); }
2290	<	catch (Exception ignore) {}
2285	>	boolean done = false;
2286	>	try {
2287	>	for (Callable<T> t : tasks) {
2288	>	ForkJoinTask<T> f = new ForkJoinTask.AdaptedCallable<T>(t);
2289	>	doSubmit(f);
2290	>	fs.add(f);
2291	>	}
2292	>	for (ForkJoinTask<T> f : fs)
2293	>	f.quietlyJoin();
2294	>	done = true;
2295	>	return futures;
2296	>	} finally {
2297	>	if (!done)
2298	>	for (ForkJoinTask<T> f : fs)
2299	>	f.cancel(false);
2300		}
2113	–	private static final long serialVersionUID = -7914297376763021607L;
2301		}
2302
2303		/**
#	Line 2175 \| Line 2362 \| public class ForkJoinPool extends Abstra
2362		int rc = 0;
2363		WorkQueue[] ws; WorkQueue w;
2364		if ((ws = workQueues) != null) {
2365	<	int n = ws.length;
2366	<	for (int i = 1; i < n; i += 2) {
2180	<	Thread.State s; ForkJoinWorkerThread wt;
2181	<	if ((w = ws[i]) != null && (wt = w.owner) != null &&
2182	<	w.eventCount >= 0 &&
2183	<	(s = wt.getState()) != Thread.State.BLOCKED &&
2184	<	s != Thread.State.WAITING &&
2185	<	s != Thread.State.TIMED_WAITING)
2365	>	for (int i = 1; i < ws.length; i += 2) {
2366	>	if ((w = ws[i]) != null && w.isApparentlyUnblocked())
2367		++rc;
2368		}
2369		}
#	Line 2231 \| Line 2412 \| public class ForkJoinPool extends Abstra
2412		long count = stealCount.get();
2413		WorkQueue[] ws; WorkQueue w;
2414		if ((ws = workQueues) != null) {
2415	<	int n = ws.length;
2235	<	for (int i = 1; i < n; i += 2) {
2415	>	for (int i = 1; i < ws.length; i += 2) {
2416		if ((w = ws[i]) != null)
2417		count += w.totalSteals;
2418		}
#	Line 2254 \| Line 2434 \| public class ForkJoinPool extends Abstra
2434		long count = 0;
2435		WorkQueue[] ws; WorkQueue w;
2436		if ((ws = workQueues) != null) {
2437	<	int n = ws.length;
2258	<	for (int i = 1; i < n; i += 2) {
2437	>	for (int i = 1; i < ws.length; i += 2) {
2438		if ((w = ws[i]) != null)
2439		count += w.queueSize();
2440		}
#	Line 2274 \| Line 2453 \| public class ForkJoinPool extends Abstra
2453		int count = 0;
2454		WorkQueue[] ws; WorkQueue w;
2455		if ((ws = workQueues) != null) {
2456	<	int n = ws.length;
2278	<	for (int i = 0; i < n; i += 2) {
2456	>	for (int i = 0; i < ws.length; i += 2) {
2457		if ((w = ws[i]) != null)
2458		count += w.queueSize();
2459		}
#	Line 2292 \| Line 2470 \| public class ForkJoinPool extends Abstra
2470		public boolean hasQueuedSubmissions() {
2471		WorkQueue[] ws; WorkQueue w;
2472		if ((ws = workQueues) != null) {
2473	<	int n = ws.length;
2474	<	for (int i = 0; i < n; i += 2) {
2297	<	if ((w = ws[i]) != null && w.queueSize() != 0)
2473	>	for (int i = 0; i < ws.length; i += 2) {
2474	>	if ((w = ws[i]) != null && !w.isEmpty())
2475		return true;
2476		}
2477		}
#	Line 2311 \| Line 2488 \| public class ForkJoinPool extends Abstra
2488		protected ForkJoinTask<?> pollSubmission() {
2489		WorkQueue[] ws; WorkQueue w; ForkJoinTask<?> t;
2490		if ((ws = workQueues) != null) {
2491	<	int n = ws.length;
2315	<	for (int i = 0; i < n; i += 2) {
2491	>	for (int i = 0; i < ws.length; i += 2) {
2492		if ((w = ws[i]) != null && (t = w.poll()) != null)
2493		return t;
2494		}
#	Line 2341 \| Line 2517 \| public class ForkJoinPool extends Abstra
2517		int count = 0;
2518		WorkQueue[] ws; WorkQueue w; ForkJoinTask<?> t;
2519		if ((ws = workQueues) != null) {
2520	<	int n = ws.length;
2345	<	for (int i = 0; i < n; ++i) {
2520	>	for (int i = 0; i < ws.length; ++i) {
2521		if ((w = ws[i]) != null) {
2522		while ((t = w.poll()) != null) {
2523		c.add(t);
#	Line 2362 \| Line 2537 \| public class ForkJoinPool extends Abstra
2537		* @return a string identifying this pool, as well as its state
2538		*/
2539		public String toString() {
2540	<	long st = getStealCount();
2541	<	long qt = getQueuedTaskCount();
2542	<	long qs = getQueuedSubmissionCount();
2368	<	int rc = getRunningThreadCount();
2369	<	int pc = parallelism;
2540	>	// Use a single pass through workQueues to collect counts
2541	>	long qt = 0L, qs = 0L; int rc = 0;
2542	>	long st = stealCount.get();
2543		long c = ctl;
2544	+	WorkQueue[] ws; WorkQueue w;
2545	+	if ((ws = workQueues) != null) {
2546	+	for (int i = 0; i < ws.length; ++i) {
2547	+	if ((w = ws[i]) != null) {
2548	+	int size = w.queueSize();
2549	+	if ((i & 1) == 0)
2550	+	qs += size;
2551	+	else {
2552	+	qt += size;
2553	+	st += w.totalSteals;
2554	+	if (w.isApparentlyUnblocked())
2555	+	++rc;
2556	+	}
2557	+	}
2558	+	}
2559	+	}
2560	+	int pc = parallelism;
2561		int tc = pc + (short)(c >>> TC_SHIFT);
2562		int ac = pc + (int)(c >> AC_SHIFT);
2563		if (ac < 0) // ignore transient negative
#	Line 2403 \| Line 2593 \| public class ForkJoinPool extends Abstra
2593		*/
2594		public void shutdown() {
2595		checkPermission();
2596	<	enableShutdown();
2407	<	tryTerminate(false);
2596	>	tryTerminate(false, true);
2597		}
2598
2599		/**
#	Line 2425 \| Line 2614 \| public class ForkJoinPool extends Abstra
2614		*/
2615		public List<Runnable> shutdownNow() {
2616		checkPermission();
2617	<	enableShutdown();
2429	<	tryTerminate(true);
2617	>	tryTerminate(true, true);
2618		return Collections.emptyList();
2619		}
2620
#	Line 2483 \| Line 2671 \| public class ForkJoinPool extends Abstra
2671		public boolean awaitTermination(long timeout, TimeUnit unit)
2672		throws InterruptedException {
2673		long nanos = unit.toNanos(timeout);
2674	<	final ReentrantLock lock = this.lock;
2674	>	final Mutex lock = this.lock;
2675		lock.lock();
2676		try {
2677		for (;;) {
#	Line 2597 \| Line 2785 \| public class ForkJoinPool extends Abstra
2785		ForkJoinPool p = ((t instanceof ForkJoinWorkerThread) ?
2786		((ForkJoinWorkerThread)t).pool : null);
2787		while (!blocker.isReleasable()) {
2788	<	if (p == null \|\| p.tryCompensate()) {
2788	>	if (p == null \|\| p.tryCompensate(null, blocker)) {
2789		try {
2790		do {} while (!blocker.isReleasable() && !blocker.block());
2791		} finally {
#	Line 2614 \| Line 2802 \| public class ForkJoinPool extends Abstra
2802		// implement RunnableFuture.
2803
2804		protected <T> RunnableFuture<T> newTaskFor(Runnable runnable, T value) {
2805	<	return (RunnableFuture<T>) ForkJoinTask.adapt(runnable, value);
2805	>	return new ForkJoinTask.AdaptedRunnable<T>(runnable, value);
2806		}
2807
2808		protected <T> RunnableFuture<T> newTaskFor(Callable<T> callable) {
2809	<	return (RunnableFuture<T>) ForkJoinTask.adapt(callable);
2809	>	return new ForkJoinTask.AdaptedCallable<T>(callable);
2810		}
2811
2812		// Unsafe mechanics
2813		private static final sun.misc.Unsafe U;
2814		private static final long CTL;
2627	–	private static final long RUNSTATE;
2815		private static final long PARKBLOCKER;
2816	+	private static final int ABASE;
2817	+	private static final int ASHIFT;
2818
2819		static {
2820		poolNumberGenerator = new AtomicInteger();
2821	+	nextSubmitterSeed = new AtomicInteger(0x55555555);
2822		modifyThreadPermission = new RuntimePermission("modifyThread");
2823		defaultForkJoinWorkerThreadFactory =
2824		new DefaultForkJoinWorkerThreadFactory();
2825	+	submitters = new ThreadSubmitter();
2826		int s;
2827		try {
2828		U = getUnsafe();
2829		Class<?> k = ForkJoinPool.class;
2830	<	Class<?> tk = Thread.class;
2830	>	Class<?> ak = ForkJoinTask[].class;
2831		CTL = U.objectFieldOffset
2832		(k.getDeclaredField("ctl"));
2833	<	RUNSTATE = U.objectFieldOffset
2643	<	(k.getDeclaredField("runState"));
2833	>	Class<?> tk = Thread.class;
2834		PARKBLOCKER = U.objectFieldOffset
2835		(tk.getDeclaredField("parkBlocker"));
2836	+	ABASE = U.arrayBaseOffset(ak);
2837	+	s = U.arrayIndexScale(ak);
2838		} catch (Exception e) {
2839		throw new Error(e);
2840		}
2841	+	if ((s & (s-1)) != 0)
2842	+	throw new Error("data type scale not a power of two");
2843	+	ASHIFT = 31 - Integer.numberOfLeadingZeros(s);
2844		}
2845
2846		/**

Diff Legend

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

Comparing jsr166/src/jsr166y/ForkJoinPool.java (file contents): Revision 1.117 by jsr166, Sat Jan 28 04:32:25 2012 UTC vs. Revision 1.127 by dl, Sun Mar 4 15:52:45 2012 UTC

Diff Legend

Comparing jsr166/src/jsr166y/ForkJoinPool.java (file contents):
Revision 1.117 by jsr166, Sat Jan 28 04:32:25 2012 UTC vs.
Revision 1.127 by dl, Sun Mar 4 15:52:45 2012 UTC