17 |
|
import java.util.concurrent.Future; |
18 |
|
import java.util.concurrent.RejectedExecutionException; |
19 |
|
import java.util.concurrent.RunnableFuture; |
20 |
– |
import java.util.concurrent.ThreadLocalRandom; |
20 |
|
import java.util.concurrent.TimeUnit; |
21 |
|
|
22 |
|
/** |
37 |
|
* ForkJoinPool}s may also be appropriate for use with event-style |
38 |
|
* tasks that are never joined. |
39 |
|
* |
40 |
< |
* <p>A static {@link #commonPool} is available and appropriate for |
40 |
> |
* <p>A static {@link #commonPool()} is available and appropriate for |
41 |
|
* most applications. The common pool is used by any ForkJoinTask that |
42 |
|
* is not explicitly submitted to a specified pool. Using the common |
43 |
|
* pool normally reduces resource usage (its threads are slowly |
51 |
|
* dynamically adding, suspending, or resuming internal worker |
52 |
|
* threads, even if some tasks are stalled waiting to join |
53 |
|
* others. However, no such adjustments are guaranteed in the face of |
54 |
< |
* blocked IO or other unmanaged synchronization. The nested {@link |
54 |
> |
* blocked I/O or other unmanaged synchronization. The nested {@link |
55 |
|
* ManagedBlocker} interface enables extension of the kinds of |
56 |
|
* synchronization accommodated. |
57 |
|
* |
99 |
|
* |
100 |
|
* <p>The common pool is by default constructed with default |
101 |
|
* parameters, but these may be controlled by setting three {@link |
102 |
< |
* System#getProperty properties} with prefix {@code |
102 |
> |
* System#getProperty system properties} with prefix {@code |
103 |
|
* java.util.concurrent.ForkJoinPool.common}: {@code parallelism} -- |
104 |
|
* an integer greater than zero, {@code threadFactory} -- the class |
105 |
|
* name of a {@link ForkJoinWorkerThreadFactory}, and {@code |
239 |
|
* enable shutdown. When used as a lock, it is normally only very |
240 |
|
* briefly held, so is nearly always available after at most a |
241 |
|
* brief spin, but we use a monitor-based backup strategy to |
242 |
< |
* blocking when needed. |
242 |
> |
* block when needed. |
243 |
|
* |
244 |
|
* Recording WorkQueues. WorkQueues are recorded in the |
245 |
|
* "workQueues" array that is created upon first use and expanded |
248 |
|
* by a lock but the array is otherwise concurrently readable, and |
249 |
|
* accessed directly. To simplify index-based operations, the |
250 |
|
* array size is always a power of two, and all readers must |
251 |
< |
* tolerate null slots. Worker queues are at odd indices Shared |
251 |
> |
* tolerate null slots. Worker queues are at odd indices. Shared |
252 |
|
* (submission) queues are at even indices, up to a maximum of 64 |
253 |
|
* slots, to limit growth even if array needs to expand to add |
254 |
|
* more workers. Grouping them together in this way simplifies and |
316 |
|
* execute. However, many other threads may notice the same task |
317 |
|
* and each signal to wake up a thread that might take it. So in |
318 |
|
* general, pools will be over-signalled. When a submission is |
319 |
< |
* added or another worker adds a task to a queue that is |
320 |
< |
* apparently empty, they signal waiting workers (or trigger |
319 |
> |
* added or another worker adds a task to a queue that has fewer |
320 |
> |
* than two tasks, they signal waiting workers (or trigger |
321 |
|
* creation of new ones if fewer than the given parallelism level |
322 |
< |
* -- see signalWork). These primary signals are buttressed by |
323 |
< |
* signals whenever other threads scan for work or do not have a |
324 |
< |
* task to process. On most platforms, signalling (unpark) |
325 |
< |
* overhead time is noticeably long, and the time between |
326 |
< |
* signalling a thread and it actually making progress can be very |
327 |
< |
* noticeably long, so it is worth offloading these delays from |
328 |
< |
* critical paths as much as possible. |
322 |
> |
* -- signalWork), and may leave a hint to the unparked worker to |
323 |
> |
* help signal others upon wakeup). These primary signals are |
324 |
> |
* buttressed by others (see method helpSignal) whenever other |
325 |
> |
* threads scan for work or do not have a task to process. On |
326 |
> |
* most platforms, signalling (unpark) overhead time is noticeably |
327 |
> |
* long, and the time between signalling a thread and it actually |
328 |
> |
* making progress can be very noticeably long, so it is worth |
329 |
> |
* offloading these delays from critical paths as much as |
330 |
> |
* possible. |
331 |
|
* |
332 |
|
* Trimming workers. To release resources after periods of lack of |
333 |
|
* use, a worker starting to wait when the pool is quiescent will |
395 |
|
* steals, rather than use per-task bookkeeping. This sometimes |
396 |
|
* requires a linear scan of workQueues array to locate stealers, |
397 |
|
* but often doesn't because stealers leave hints (that may become |
398 |
< |
* stale/wrong) of where to locate them. A stealHint is only a |
399 |
< |
* hint because a worker might have had multiple steals and the |
400 |
< |
* hint records only one of them (usually the most current). |
401 |
< |
* Hinting isolates cost to when it is needed, rather than adding |
402 |
< |
* to per-task overhead. (2) It is "shallow", ignoring nesting |
403 |
< |
* and potentially cyclic mutual steals. (3) It is intentionally |
398 |
> |
* stale/wrong) of where to locate them. It is only a hint |
399 |
> |
* because a worker might have had multiple steals and the hint |
400 |
> |
* records only one of them (usually the most current). Hinting |
401 |
> |
* isolates cost to when it is needed, rather than adding to |
402 |
> |
* per-task overhead. (2) It is "shallow", ignoring nesting and |
403 |
> |
* potentially cyclic mutual steals. (3) It is intentionally |
404 |
|
* racy: field currentJoin is updated only while actively joining, |
405 |
|
* which means that we miss links in the chain during long-lived |
406 |
|
* tasks, GC stalls etc (which is OK since blocking in such cases |
440 |
|
* Common Pool |
441 |
|
* =========== |
442 |
|
* |
443 |
< |
* The static commonPool always exists after static |
443 |
> |
* The static common Pool always exists after static |
444 |
|
* initialization. Since it (or any other created pool) need |
445 |
|
* never be used, we minimize initial construction overhead and |
446 |
|
* footprint to the setup of about a dozen fields, with no nested |
527 |
|
* Default ForkJoinWorkerThreadFactory implementation; creates a |
528 |
|
* new ForkJoinWorkerThread. |
529 |
|
*/ |
530 |
< |
static class DefaultForkJoinWorkerThreadFactory |
530 |
> |
static final class DefaultForkJoinWorkerThreadFactory |
531 |
|
implements ForkJoinWorkerThreadFactory { |
532 |
< |
public ForkJoinWorkerThread newThread(ForkJoinPool pool) { |
532 |
> |
public final ForkJoinWorkerThread newThread(ForkJoinPool pool) { |
533 |
|
return new ForkJoinWorkerThread(pool); |
534 |
|
} |
535 |
|
} |
536 |
|
|
537 |
|
/** |
538 |
+ |
* Per-thread records for threads that submit to pools. Currently |
539 |
+ |
* holds only pseudo-random seed / index that is used to choose |
540 |
+ |
* submission queues in method externalPush. In the future, this may |
541 |
+ |
* also incorporate a means to implement different task rejection |
542 |
+ |
* and resubmission policies. |
543 |
+ |
* |
544 |
+ |
* Seeds for submitters and workers/workQueues work in basically |
545 |
+ |
* the same way but are initialized and updated using slightly |
546 |
+ |
* different mechanics. Both are initialized using the same |
547 |
+ |
* approach as in class ThreadLocal, where successive values are |
548 |
+ |
* unlikely to collide with previous values. Seeds are then |
549 |
+ |
* randomly modified upon collisions using xorshifts, which |
550 |
+ |
* requires a non-zero seed. |
551 |
+ |
*/ |
552 |
+ |
static final class Submitter { |
553 |
+ |
int seed; |
554 |
+ |
Submitter(int s) { seed = s; } |
555 |
+ |
} |
556 |
+ |
|
557 |
+ |
/** |
558 |
|
* Class for artificial tasks that are used to replace the target |
559 |
|
* of local joins if they are removed from an interior queue slot |
560 |
|
* in WorkQueue.tryRemoveAndExec. We don't need the proxy to |
620 |
|
* trades off slightly slower average field access for the sake of |
621 |
|
* avoiding really bad worst-case access. (Until better JVM |
622 |
|
* support is in place, this padding is dependent on transient |
623 |
< |
* properties of JVM field layout rules.) We also take care in |
623 |
> |
* properties of JVM field layout rules.) We also take care in |
624 |
|
* allocating, sizing and resizing the array. Non-shared queue |
625 |
|
* arrays are initialized by workers before use. Others are |
626 |
|
* allocated on first use. |
646 |
|
*/ |
647 |
|
static final int MAXIMUM_QUEUE_CAPACITY = 1 << 26; // 64M |
648 |
|
|
649 |
+ |
// Heuristic padding to ameliorate unfortunate memory placements |
650 |
+ |
volatile long pad00, pad01, pad02, pad03, pad04, pad05, pad06; |
651 |
+ |
|
652 |
|
int seed; // for random scanning; initialize nonzero |
653 |
|
volatile int eventCount; // encoded inactivation count; < 0 if inactive |
654 |
|
int nextWait; // encoded record of next event waiter |
655 |
< |
final int mode; // lifo, fifo, or shared |
632 |
< |
int nsteals; // cumulative number of steals |
655 |
> |
int hint; // steal or signal hint (index) |
656 |
|
int poolIndex; // index of this queue in pool (or 0) |
657 |
< |
int stealHint; // index of most recent known stealer |
657 |
> |
final int mode; // 0: lifo, > 0: fifo, < 0: shared |
658 |
> |
int nsteals; // number of steals |
659 |
|
volatile int qlock; // 1: locked, -1: terminate; else 0 |
660 |
|
volatile int base; // index of next slot for poll |
661 |
|
int top; // index of next slot for push |
665 |
|
volatile Thread parker; // == owner during call to park; else null |
666 |
|
volatile ForkJoinTask<?> currentJoin; // task being joined in awaitJoin |
667 |
|
ForkJoinTask<?> currentSteal; // current non-local task being executed |
644 |
– |
// Heuristic padding to ameliorate unfortunate memory placements |
645 |
– |
Object p00, p01, p02, p03, p04, p05, p06, p07; |
646 |
– |
Object p08, p09, p0a, p0b, p0c, p0d, p0e; |
668 |
|
|
669 |
< |
WorkQueue(ForkJoinPool pool, ForkJoinWorkerThread owner, int mode) { |
670 |
< |
this.mode = mode; |
669 |
> |
volatile Object pad10, pad11, pad12, pad13, pad14, pad15, pad16, pad17; |
670 |
> |
volatile Object pad18, pad19, pad1a, pad1b, pad1c, pad1d; |
671 |
> |
|
672 |
> |
WorkQueue(ForkJoinPool pool, ForkJoinWorkerThread owner, int mode, |
673 |
> |
int seed) { |
674 |
|
this.pool = pool; |
675 |
|
this.owner = owner; |
676 |
+ |
this.mode = mode; |
677 |
+ |
this.seed = seed; |
678 |
|
// Place indices in the center of array (that is not yet allocated) |
679 |
|
base = top = INITIAL_QUEUE_CAPACITY >>> 1; |
680 |
|
} |
681 |
|
|
682 |
|
/** |
683 |
< |
* Pushes a task. Call only by owner in unshared queues. |
658 |
< |
* Cases needing resizing or rejection are relayed to fullPush |
659 |
< |
* (that also handles shared queues). |
660 |
< |
* |
661 |
< |
* @param task the task. Caller must ensure non-null. |
662 |
< |
* @throw RejectedExecutionException if array cannot be resized |
683 |
> |
* Returns the approximate number of tasks in the queue. |
684 |
|
*/ |
685 |
< |
final void push(ForkJoinTask<?> task) { |
686 |
< |
ForkJoinPool p; ForkJoinTask<?>[] a; |
687 |
< |
int s = top, n; |
688 |
< |
if ((a = array) != null && a.length > (n = s + 1 - base)) { |
689 |
< |
U.putOrderedObject |
690 |
< |
(a, (((a.length - 1) & s) << ASHIFT) + ABASE, task); |
691 |
< |
top = s + 1; |
692 |
< |
if (n <= 1 && (p = pool) != null) |
693 |
< |
p.signalWork(this, 1); |
694 |
< |
} |
695 |
< |
else |
696 |
< |
fullPush(task, true); |
685 |
> |
final int queueSize() { |
686 |
> |
int n = base - top; // non-owner callers must read base first |
687 |
> |
return (n >= 0) ? 0 : -n; // ignore transient negative |
688 |
> |
} |
689 |
> |
|
690 |
> |
/** |
691 |
> |
* Provides a more accurate estimate of whether this queue has |
692 |
> |
* any tasks than does queueSize, by checking whether a |
693 |
> |
* near-empty queue has at least one unclaimed task. |
694 |
> |
*/ |
695 |
> |
final boolean isEmpty() { |
696 |
> |
ForkJoinTask<?>[] a; int m, s; |
697 |
> |
int n = base - (s = top); |
698 |
> |
return (n >= 0 || |
699 |
> |
(n == -1 && |
700 |
> |
((a = array) == null || |
701 |
> |
(m = a.length - 1) < 0 || |
702 |
> |
U.getObject |
703 |
> |
(a, (long)((m & (s - 1)) << ASHIFT) + ABASE) == null))); |
704 |
|
} |
705 |
|
|
706 |
|
/** |
707 |
< |
* Pushes a task if lock is free and array is either big |
708 |
< |
* enough or can be resized to be big enough. Note: a |
681 |
< |
* specialization of a common fast path of this method is in |
682 |
< |
* ForkJoinPool.externalPush. When called from a FJWT queue, |
683 |
< |
* this can fail only if the pool has been shut down or |
684 |
< |
* an out of memory error. |
707 |
> |
* Pushes a task. Call only by owner in unshared queues. (The |
708 |
> |
* shared-queue version is embedded in method externalPush.) |
709 |
|
* |
710 |
|
* @param task the task. Caller must ensure non-null. |
711 |
< |
* @param owned if true, throw RJE on failure |
711 |
> |
* @throw RejectedExecutionException if array cannot be resized |
712 |
|
*/ |
713 |
< |
final boolean fullPush(ForkJoinTask<?> task, boolean owned) { |
714 |
< |
ForkJoinPool p; ForkJoinTask<?>[] a; |
715 |
< |
if (owned) { |
716 |
< |
if (qlock < 0) // must be shutting down |
717 |
< |
throw new RejectedExecutionException(); |
718 |
< |
} |
719 |
< |
else if (!U.compareAndSwapInt(this, QLOCK, 0, 1)) |
720 |
< |
return false; |
721 |
< |
try { |
698 |
< |
int s = top, oldLen, len; |
699 |
< |
if ((a = array) == null) |
700 |
< |
a = array = new ForkJoinTask<?>[len=INITIAL_QUEUE_CAPACITY]; |
701 |
< |
else if ((oldLen = a.length) > s + 1 - base) |
702 |
< |
len = oldLen; |
703 |
< |
else if ((len = oldLen << 1) > MAXIMUM_QUEUE_CAPACITY) |
704 |
< |
throw new RejectedExecutionException("Capacity exceeded"); |
705 |
< |
else { |
706 |
< |
int oldMask, b; |
707 |
< |
ForkJoinTask<?>[] oldA = a; |
708 |
< |
a = array = new ForkJoinTask<?>[len]; |
709 |
< |
if ((oldMask = oldLen - 1) >= 0 && s - (b = base) > 0) { |
710 |
< |
int mask = len - 1; |
711 |
< |
do { |
712 |
< |
ForkJoinTask<?> x; |
713 |
< |
int oldj = ((b & oldMask) << ASHIFT) + ABASE; |
714 |
< |
int j = ((b & mask) << ASHIFT) + ABASE; |
715 |
< |
x = (ForkJoinTask<?>) |
716 |
< |
U.getObjectVolatile(oldA, oldj); |
717 |
< |
if (x != null && |
718 |
< |
U.compareAndSwapObject(oldA, oldj, x, null)) |
719 |
< |
U.putObjectVolatile(a, j, x); |
720 |
< |
} while (++b != s); |
721 |
< |
} |
713 |
> |
final void push(ForkJoinTask<?> task) { |
714 |
> |
ForkJoinTask<?>[] a; ForkJoinPool p; |
715 |
> |
int s = top, m, n; |
716 |
> |
if ((a = array) != null) { // ignore if queue removed |
717 |
> |
int j = (((m = a.length - 1) & s) << ASHIFT) + ABASE; |
718 |
> |
U.putOrderedObject(a, j, task); |
719 |
> |
if ((n = (top = s + 1) - base) <= 2) { |
720 |
> |
if ((p = pool) != null) |
721 |
> |
p.signalWork(this); |
722 |
|
} |
723 |
< |
U.putOrderedObject |
724 |
< |
(a, (((len - 1) & s) << ASHIFT) + ABASE, task); |
725 |
< |
top = s + 1; |
726 |
< |
} finally { |
727 |
< |
if (!owned) |
728 |
< |
qlock = 0; |
723 |
> |
else if (n >= m) |
724 |
> |
growArray(); |
725 |
|
} |
726 |
< |
if ((p = pool) != null) |
727 |
< |
p.signalWork(this, 1); |
728 |
< |
return true; |
726 |
> |
} |
727 |
> |
|
728 |
> |
/** |
729 |
> |
* Initializes or doubles the capacity of array. Call either |
730 |
> |
* by owner or with lock held -- it is OK for base, but not |
731 |
> |
* top, to move while resizings are in progress. |
732 |
> |
*/ |
733 |
> |
final ForkJoinTask<?>[] growArray() { |
734 |
> |
ForkJoinTask<?>[] oldA = array; |
735 |
> |
int size = oldA != null ? oldA.length << 1 : INITIAL_QUEUE_CAPACITY; |
736 |
> |
if (size > MAXIMUM_QUEUE_CAPACITY) |
737 |
> |
throw new RejectedExecutionException("Queue capacity exceeded"); |
738 |
> |
int oldMask, t, b; |
739 |
> |
ForkJoinTask<?>[] a = array = new ForkJoinTask<?>[size]; |
740 |
> |
if (oldA != null && (oldMask = oldA.length - 1) >= 0 && |
741 |
> |
(t = top) - (b = base) > 0) { |
742 |
> |
int mask = size - 1; |
743 |
> |
do { |
744 |
> |
ForkJoinTask<?> x; |
745 |
> |
int oldj = ((b & oldMask) << ASHIFT) + ABASE; |
746 |
> |
int j = ((b & mask) << ASHIFT) + ABASE; |
747 |
> |
x = (ForkJoinTask<?>)U.getObjectVolatile(oldA, oldj); |
748 |
> |
if (x != null && |
749 |
> |
U.compareAndSwapObject(oldA, oldj, x, null)) |
750 |
> |
U.putObjectVolatile(a, j, x); |
751 |
> |
} while (++b != t); |
752 |
> |
} |
753 |
> |
return a; |
754 |
|
} |
755 |
|
|
756 |
|
/** |
874 |
|
return seed = r ^= r << 5; |
875 |
|
} |
876 |
|
|
856 |
– |
/** |
857 |
– |
* Provides a more accurate estimate of size than (top - base) |
858 |
– |
* by ordering reads and checking whether a near-empty queue |
859 |
– |
* has at least one unclaimed task. |
860 |
– |
*/ |
861 |
– |
final int queueSize() { |
862 |
– |
ForkJoinTask<?>[] a; int k, s, n; |
863 |
– |
return ((n = base - (s = top)) < 0 && |
864 |
– |
(n != -1 || |
865 |
– |
((a = array) != null && (k = a.length) > 0 && |
866 |
– |
U.getObject |
867 |
– |
(a, (long)((((k - 1) & (s - 1)) << ASHIFT) + ABASE)) != null))) ? |
868 |
– |
-n : 0; |
869 |
– |
} |
870 |
– |
|
877 |
|
// Specialized execution methods |
878 |
|
|
879 |
|
/** |
987 |
|
if (t != null) { |
988 |
|
(currentSteal = t).doExec(); |
989 |
|
currentSteal = null; |
990 |
< |
if (++nsteals < 0) { // spill on overflow |
991 |
< |
ForkJoinPool p; |
986 |
< |
if ((p = pool) != null) |
987 |
< |
p.collectStealCount(this); |
988 |
< |
} |
989 |
< |
if (top != base) { // process remaining local tasks |
990 |
> |
++nsteals; |
991 |
> |
if (base - top < 0) { // process remaining local tasks |
992 |
|
if (mode == 0) |
993 |
|
popAndExecAll(); |
994 |
|
else |
1020 |
|
s != Thread.State.TIMED_WAITING); |
1021 |
|
} |
1022 |
|
|
1021 |
– |
/** |
1022 |
– |
* If this owned and is not already interrupted, try to |
1023 |
– |
* interrupt and/or unpark, ignoring exceptions. |
1024 |
– |
*/ |
1025 |
– |
final void interruptOwner() { |
1026 |
– |
Thread wt, p; |
1027 |
– |
if ((wt = owner) != null && !wt.isInterrupted()) { |
1028 |
– |
try { |
1029 |
– |
wt.interrupt(); |
1030 |
– |
} catch (SecurityException ignore) { |
1031 |
– |
} |
1032 |
– |
} |
1033 |
– |
if ((p = parker) != null) |
1034 |
– |
U.unpark(p); |
1035 |
– |
} |
1036 |
– |
|
1023 |
|
// Unsafe mechanics |
1024 |
|
private static final sun.misc.Unsafe U; |
1025 |
|
private static final long QLOCK; |
1044 |
|
} |
1045 |
|
} |
1046 |
|
|
1061 |
– |
/** |
1062 |
– |
* Per-thread records for threads that submit to pools. Currently |
1063 |
– |
* holds only pseudo-random seed / index that is used to choose |
1064 |
– |
* submission queues in method externalPush. In the future, this may |
1065 |
– |
* also incorporate a means to implement different task rejection |
1066 |
– |
* and resubmission policies. |
1067 |
– |
* |
1068 |
– |
* Seeds for submitters and workers/workQueues work in basically |
1069 |
– |
* the same way but are initialized and updated using slightly |
1070 |
– |
* different mechanics. Both are initialized using the same |
1071 |
– |
* approach as in class ThreadLocal, where successive values are |
1072 |
– |
* unlikely to collide with previous values. Seeds are then |
1073 |
– |
* randomly modified upon collisions using xorshifts, which |
1074 |
– |
* requires a non-zero seed. |
1075 |
– |
*/ |
1076 |
– |
static final class Submitter { |
1077 |
– |
int seed; |
1078 |
– |
Submitter(int s) { seed = s; } |
1079 |
– |
} |
1080 |
– |
|
1081 |
– |
/** Property prefix for constructing common pool */ |
1082 |
– |
private static final String propPrefix = |
1083 |
– |
"java.util.concurrent.ForkJoinPool.common."; |
1084 |
– |
|
1047 |
|
// static fields (initialized in static initializer below) |
1048 |
|
|
1049 |
|
/** |
1054 |
|
defaultForkJoinWorkerThreadFactory; |
1055 |
|
|
1056 |
|
/** |
1057 |
< |
* Common (static) pool. Non-null for public use unless a static |
1058 |
< |
* construction exception, but internal usages null-check on use |
1059 |
< |
* to paranoically avoid potential initialization circularities |
1060 |
< |
* as well as to simplify generated code. |
1057 |
> |
* Per-thread submission bookkeeping. Shared across all pools |
1058 |
> |
* to reduce ThreadLocal pollution and because random motion |
1059 |
> |
* to avoid contention in one pool is likely to hold for others. |
1060 |
> |
* Lazily initialized on first submission (but null-checked |
1061 |
> |
* in other contexts to avoid unnecessary initialization). |
1062 |
|
*/ |
1063 |
< |
static final ForkJoinPool commonPool; |
1063 |
> |
static final ThreadLocal<Submitter> submitters; |
1064 |
|
|
1065 |
|
/** |
1066 |
|
* Permission required for callers of methods that may start or |
1069 |
|
private static final RuntimePermission modifyThreadPermission; |
1070 |
|
|
1071 |
|
/** |
1072 |
< |
* Per-thread submission bookkeeping. Shared across all pools |
1073 |
< |
* to reduce ThreadLocal pollution and because random motion |
1074 |
< |
* to avoid contention in one pool is likely to hold for others. |
1075 |
< |
* Lazily initialized on first submission (but null-checked |
1113 |
< |
* in other contexts to avoid unnecessary initialization). |
1072 |
> |
* Common (static) pool. Non-null for public use unless a static |
1073 |
> |
* construction exception, but internal usages null-check on use |
1074 |
> |
* to paranoically avoid potential initialization circularities |
1075 |
> |
* as well as to simplify generated code. |
1076 |
|
*/ |
1077 |
< |
static final ThreadLocal<Submitter> submitters; |
1077 |
> |
static final ForkJoinPool common; |
1078 |
|
|
1079 |
|
/** |
1080 |
< |
* Common pool parallelism. Must equal commonPool.parallelism. |
1080 |
> |
* Common pool parallelism. Must equal common.parallelism. |
1081 |
|
*/ |
1082 |
< |
static final int commonPoolParallelism; |
1082 |
> |
static final int commonParallelism; |
1083 |
|
|
1084 |
|
/** |
1085 |
|
* Sequence number for creating workerNamePrefix. |
1087 |
|
private static int poolNumberSequence; |
1088 |
|
|
1089 |
|
/** |
1090 |
< |
* Return the next sequence number. We don't expect this to |
1091 |
< |
* ever contend so use simple builtin sync. |
1090 |
> |
* Returns the next sequence number. We don't expect this to |
1091 |
> |
* ever contend, so use simple builtin sync. |
1092 |
|
*/ |
1093 |
|
private static final synchronized int nextPoolId() { |
1094 |
|
return ++poolNumberSequence; |
1112 |
|
private static final long FAST_IDLE_TIMEOUT = 200L * 1000L * 1000L; |
1113 |
|
|
1114 |
|
/** |
1115 |
+ |
* Tolerance for idle timeouts, to cope with timer undershoots |
1116 |
+ |
*/ |
1117 |
+ |
private static final long TIMEOUT_SLOP = 2000000L; |
1118 |
+ |
|
1119 |
+ |
/** |
1120 |
|
* The maximum stolen->joining link depth allowed in method |
1121 |
|
* tryHelpStealer. Must be a power of two. Depths for legitimate |
1122 |
|
* chains are unbounded, but we use a fixed constant to avoid |
1216 |
|
static final int FIFO_QUEUE = 1; |
1217 |
|
static final int SHARED_QUEUE = -1; |
1218 |
|
|
1219 |
+ |
// bounds for #steps in scan loop -- must be power 2 minus 1 |
1220 |
+ |
private static final int MIN_SCAN = 0x1ff; // cover estimation slop |
1221 |
+ |
private static final int MAX_SCAN = 0x1ffff; // 4 * max workers |
1222 |
+ |
|
1223 |
|
// Instance fields |
1224 |
|
|
1225 |
|
/* |
1226 |
< |
* Field layout order in this class tends to matter more than one |
1227 |
< |
* would like. Runtime layout order is only loosely related to |
1226 |
> |
* Field layout of this class tends to matter more than one would |
1227 |
> |
* like. Runtime layout order is only loosely related to |
1228 |
|
* declaration order and may differ across JVMs, but the following |
1229 |
|
* empirically works OK on current JVMs. |
1230 |
|
*/ |
1231 |
+ |
|
1232 |
+ |
// Heuristic padding to ameliorate unfortunate memory placements |
1233 |
+ |
volatile long pad00, pad01, pad02, pad03, pad04, pad05, pad06; |
1234 |
+ |
|
1235 |
|
volatile long stealCount; // collects worker counts |
1236 |
|
volatile long ctl; // main pool control |
1262 |
– |
final int parallelism; // parallelism level |
1263 |
– |
final int localMode; // per-worker scheduling mode |
1264 |
– |
volatile int indexSeed; // worker/submitter index seed |
1237 |
|
volatile int plock; // shutdown status and seqLock |
1238 |
+ |
volatile int indexSeed; // worker/submitter index seed |
1239 |
+ |
final int config; // mode and parallelism level |
1240 |
|
WorkQueue[] workQueues; // main registry |
1241 |
< |
final ForkJoinWorkerThreadFactory factory; // factory for new workers |
1241 |
> |
final ForkJoinWorkerThreadFactory factory; |
1242 |
|
final Thread.UncaughtExceptionHandler ueh; // per-worker UEH |
1243 |
|
final String workerNamePrefix; // to create worker name string |
1244 |
|
|
1245 |
+ |
volatile Object pad10, pad11, pad12, pad13, pad14, pad15, pad16, pad17; |
1246 |
+ |
volatile Object pad18, pad19, pad1a, pad1b; |
1247 |
+ |
|
1248 |
|
/* |
1249 |
|
* Acquires the plock lock to protect worker array and related |
1250 |
|
* updates. This method is called only if an initial CAS on plock |
1251 |
|
* fails. This acts as a spinLock for normal cases, but falls back |
1252 |
|
* to builtin monitor to block when (rarely) needed. This would be |
1253 |
|
* a terrible idea for a highly contended lock, but works fine as |
1254 |
< |
* a more conservative alternative to a pure spinlock. See |
1278 |
< |
* internal ConcurrentHashMap documentation for further |
1279 |
< |
* explanation of nearly the same construction. |
1254 |
> |
* a more conservative alternative to a pure spinlock. |
1255 |
|
*/ |
1256 |
|
private int acquirePlock() { |
1257 |
|
int spins = PL_SPINS, r = 0, ps, nps; |
1259 |
|
if (((ps = plock) & PL_LOCK) == 0 && |
1260 |
|
U.compareAndSwapInt(this, PLOCK, ps, nps = ps + PL_LOCK)) |
1261 |
|
return nps; |
1262 |
< |
else if (r == 0) |
1263 |
< |
r = ThreadLocalRandom.current().nextInt(); // randomize spins |
1262 |
> |
else if (r == 0) { // randomize spins if possible |
1263 |
> |
Thread t = Thread.currentThread(); WorkQueue w; Submitter z; |
1264 |
> |
if ((t instanceof ForkJoinWorkerThread) && |
1265 |
> |
(w = ((ForkJoinWorkerThread)t).workQueue) != null) |
1266 |
> |
r = w.seed; |
1267 |
> |
else if ((z = submitters.get()) != null) |
1268 |
> |
r = z.seed; |
1269 |
> |
else |
1270 |
> |
r = 1; |
1271 |
> |
} |
1272 |
|
else if (spins >= 0) { |
1273 |
|
r ^= r << 1; r ^= r >>> 3; r ^= r << 10; // xorshift |
1274 |
|
if (r >= 0) |
1302 |
|
synchronized (this) { notifyAll(); } |
1303 |
|
} |
1304 |
|
|
1305 |
+ |
/** |
1306 |
+ |
* Tries to create and start one worker if fewer than target |
1307 |
+ |
* parallelism level exist. Adjusts counts etc on failure. |
1308 |
+ |
*/ |
1309 |
+ |
private void tryAddWorker() { |
1310 |
+ |
long c; int u; |
1311 |
+ |
while ((u = (int)((c = ctl) >>> 32)) < 0 && |
1312 |
+ |
(u & SHORT_SIGN) != 0 && (int)c == 0) { |
1313 |
+ |
long nc = (long)(((u + UTC_UNIT) & UTC_MASK) | |
1314 |
+ |
((u + UAC_UNIT) & UAC_MASK)) << 32; |
1315 |
+ |
if (U.compareAndSwapLong(this, CTL, c, nc)) { |
1316 |
+ |
ForkJoinWorkerThreadFactory fac; |
1317 |
+ |
Throwable ex = null; |
1318 |
+ |
ForkJoinWorkerThread wt = null; |
1319 |
+ |
try { |
1320 |
+ |
if ((fac = factory) != null && |
1321 |
+ |
(wt = fac.newThread(this)) != null) { |
1322 |
+ |
wt.start(); |
1323 |
+ |
break; |
1324 |
+ |
} |
1325 |
+ |
} catch (Throwable e) { |
1326 |
+ |
ex = e; |
1327 |
+ |
} |
1328 |
+ |
deregisterWorker(wt, ex); |
1329 |
+ |
break; |
1330 |
+ |
} |
1331 |
+ |
} |
1332 |
+ |
} |
1333 |
+ |
|
1334 |
|
// Registering and deregistering workers |
1335 |
|
|
1336 |
|
/** |
1337 |
< |
* Callback from ForkJoinWorkerThread constructor to establish its |
1338 |
< |
* poolIndex and record its WorkQueue. To avoid scanning bias due |
1339 |
< |
* to packing entries in front of the workQueues array, we treat |
1340 |
< |
* the array as a simple power-of-two hash table using per-thread |
1341 |
< |
* seed as hash, expanding as needed. |
1342 |
< |
* |
1343 |
< |
* @param w the worker's queue |
1344 |
< |
*/ |
1345 |
< |
final void registerWorker(WorkQueue w) { |
1346 |
< |
int s, ps; // generate a rarely colliding candidate index seed |
1347 |
< |
do {} while (!U.compareAndSwapInt(this, INDEXSEED, |
1348 |
< |
s = indexSeed, s += SEED_INCREMENT) || |
1337 |
> |
* Callback from ForkJoinWorkerThread to establish and record its |
1338 |
> |
* WorkQueue. To avoid scanning bias due to packing entries in |
1339 |
> |
* front of the workQueues array, we treat the array as a simple |
1340 |
> |
* power-of-two hash table using per-thread seed as hash, |
1341 |
> |
* expanding as needed. |
1342 |
> |
* |
1343 |
> |
* @param wt the worker thread |
1344 |
> |
* @return the worker's queue |
1345 |
> |
*/ |
1346 |
> |
final WorkQueue registerWorker(ForkJoinWorkerThread wt) { |
1347 |
> |
Thread.UncaughtExceptionHandler handler; WorkQueue[] ws; int s, ps; |
1348 |
> |
wt.setDaemon(true); |
1349 |
> |
if ((handler = ueh) != null) |
1350 |
> |
wt.setUncaughtExceptionHandler(handler); |
1351 |
> |
do {} while (!U.compareAndSwapInt(this, INDEXSEED, s = indexSeed, |
1352 |
> |
s += SEED_INCREMENT) || |
1353 |
|
s == 0); // skip 0 |
1354 |
+ |
WorkQueue w = new WorkQueue(this, wt, config >>> 16, s); |
1355 |
|
if (((ps = plock) & PL_LOCK) != 0 || |
1356 |
|
!U.compareAndSwapInt(this, PLOCK, ps, ps += PL_LOCK)) |
1357 |
|
ps = acquirePlock(); |
1358 |
|
int nps = (ps & SHUTDOWN) | ((ps + PL_LOCK) & ~SHUTDOWN); |
1359 |
|
try { |
1360 |
< |
WorkQueue[] ws; |
1344 |
< |
if (w != null && (ws = workQueues) != null) { |
1345 |
< |
w.seed = s; |
1360 |
> |
if ((ws = workQueues) != null) { // skip if shutting down |
1361 |
|
int n = ws.length, m = n - 1; |
1362 |
< |
int r = (s << 1) | 1; // use odd-numbered indices |
1363 |
< |
if (ws[r &= m] != null) { // collision |
1364 |
< |
int probes = 0; // step by approx half size |
1362 |
> |
int r = (s << 1) | 1; // use odd-numbered indices |
1363 |
> |
if (ws[r &= m] != null) { // collision |
1364 |
> |
int probes = 0; // step by approx half size |
1365 |
|
int step = (n <= 4) ? 2 : ((n >>> 1) & EVENMASK) + 2; |
1366 |
|
while (ws[r = (r + step) & m] != null) { |
1367 |
|
if (++probes >= n) { |
1371 |
|
} |
1372 |
|
} |
1373 |
|
} |
1374 |
< |
w.eventCount = w.poolIndex = r; // establish before recording |
1374 |
> |
w.eventCount = w.poolIndex = r; // volatile write orders |
1375 |
|
ws[r] = w; |
1376 |
|
} |
1377 |
|
} finally { |
1378 |
|
if (!U.compareAndSwapInt(this, PLOCK, ps, nps)) |
1379 |
|
releasePlock(nps); |
1380 |
|
} |
1381 |
+ |
wt.setName(workerNamePrefix.concat(Integer.toString(w.poolIndex))); |
1382 |
+ |
return w; |
1383 |
|
} |
1384 |
|
|
1385 |
|
/** |
1395 |
|
WorkQueue w = null; |
1396 |
|
if (wt != null && (w = wt.workQueue) != null) { |
1397 |
|
int ps; |
1381 |
– |
collectStealCount(w); |
1398 |
|
w.qlock = -1; // ensure set |
1399 |
+ |
long ns = w.nsteals, sc; // collect steal count |
1400 |
+ |
do {} while (!U.compareAndSwapLong(this, STEALCOUNT, |
1401 |
+ |
sc = stealCount, sc + ns)); |
1402 |
|
if (((ps = plock) & PL_LOCK) != 0 || |
1403 |
|
!U.compareAndSwapInt(this, PLOCK, ps, ps += PL_LOCK)) |
1404 |
|
ps = acquirePlock(); |
1414 |
|
} |
1415 |
|
} |
1416 |
|
|
1417 |
< |
long c; // adjust ctl counts |
1417 |
> |
long c; // adjust ctl counts |
1418 |
|
do {} while (!U.compareAndSwapLong |
1419 |
|
(this, CTL, c = ctl, (((c - AC_UNIT) & AC_MASK) | |
1420 |
|
((c - TC_UNIT) & TC_MASK) | |
1421 |
|
(c & ~(AC_MASK|TC_MASK))))); |
1422 |
|
|
1423 |
< |
if (!tryTerminate(false, false) && w != null) { |
1424 |
< |
w.cancelAll(); // cancel remaining tasks |
1425 |
< |
if (w.array != null) // suppress signal if never ran |
1426 |
< |
signalWork(null, 1); // wake up or create replacement |
1427 |
< |
if (ex == null) // help clean refs on way out |
1428 |
< |
ForkJoinTask.helpExpungeStaleExceptions(); |
1423 |
> |
if (!tryTerminate(false, false) && w != null && w.array != null) { |
1424 |
> |
w.cancelAll(); // cancel remaining tasks |
1425 |
> |
WorkQueue[] ws; WorkQueue v; Thread p; int u, i, e; |
1426 |
> |
while ((u = (int)((c = ctl) >>> 32)) < 0 && (e = (int)c) >= 0) { |
1427 |
> |
if (e > 0) { // activate or create replacement |
1428 |
> |
if ((ws = workQueues) == null || |
1429 |
> |
(i = e & SMASK) >= ws.length || |
1430 |
> |
(v = ws[i]) == null) |
1431 |
> |
break; |
1432 |
> |
long nc = (((long)(v.nextWait & E_MASK)) | |
1433 |
> |
((long)(u + UAC_UNIT) << 32)); |
1434 |
> |
if (v.eventCount != (e | INT_SIGN)) |
1435 |
> |
break; |
1436 |
> |
if (U.compareAndSwapLong(this, CTL, c, nc)) { |
1437 |
> |
v.eventCount = (e + E_SEQ) & E_MASK; |
1438 |
> |
if ((p = v.parker) != null) |
1439 |
> |
U.unpark(p); |
1440 |
> |
break; |
1441 |
> |
} |
1442 |
> |
} |
1443 |
> |
else { |
1444 |
> |
if ((short)u < 0) |
1445 |
> |
tryAddWorker(); |
1446 |
> |
break; |
1447 |
> |
} |
1448 |
> |
} |
1449 |
|
} |
1450 |
< |
|
1451 |
< |
if (ex != null) // rethrow |
1450 |
> |
if (ex == null) // help clean refs on way out |
1451 |
> |
ForkJoinTask.helpExpungeStaleExceptions(); |
1452 |
> |
else // rethrow |
1453 |
|
ForkJoinTask.rethrow(ex); |
1454 |
|
} |
1455 |
|
|
1416 |
– |
/** |
1417 |
– |
* Collect worker steal count into total. Called on termination |
1418 |
– |
* and upon int overflow of local count. (There is a possible race |
1419 |
– |
* in the latter case vs any caller of getStealCount, which can |
1420 |
– |
* make its results less accurate than usual.) |
1421 |
– |
*/ |
1422 |
– |
final void collectStealCount(WorkQueue w) { |
1423 |
– |
if (w != null) { |
1424 |
– |
long sc; |
1425 |
– |
int ns = w.nsteals; |
1426 |
– |
w.nsteals = 0; // handle overflow |
1427 |
– |
long steals = (ns >= 0) ? ns : 1L + (long)(Integer.MAX_VALUE); |
1428 |
– |
do {} while (!U.compareAndSwapLong(this, STEALCOUNT, |
1429 |
– |
sc = stealCount, sc + steals)); |
1430 |
– |
} |
1431 |
– |
} |
1432 |
– |
|
1456 |
|
// Submissions |
1457 |
|
|
1458 |
|
/** |
1469 |
|
(ws = workQueues) != null && (m = (ws.length - 1)) >= 0 && |
1470 |
|
(q = ws[m & z.seed & SQMASK]) != null && |
1471 |
|
U.compareAndSwapInt(q, QLOCK, 0, 1)) { // lock |
1472 |
< |
int s = q.top, n; |
1473 |
< |
if ((a = q.array) != null && a.length > (n = s + 1 - q.base)) { |
1474 |
< |
U.putObject(a, (long)(((a.length - 1) & s) << ASHIFT) + ABASE, |
1475 |
< |
task); |
1472 |
> |
int b = q.base, s = q.top, n, an; |
1473 |
> |
if ((a = q.array) != null && (an = a.length) > (n = s + 1 - b)) { |
1474 |
> |
int j = (((an - 1) & s) << ASHIFT) + ABASE; |
1475 |
> |
U.putOrderedObject(a, j, task); |
1476 |
|
q.top = s + 1; // push on to deque |
1477 |
|
q.qlock = 0; |
1478 |
< |
if (n <= 1) |
1479 |
< |
signalWork(q, 1); |
1478 |
> |
if (n <= 2) |
1479 |
> |
signalWork(q); |
1480 |
|
return; |
1481 |
|
} |
1482 |
|
q.qlock = 0; |
1487 |
|
/** |
1488 |
|
* Full version of externalPush. This method is called, among |
1489 |
|
* other times, upon the first submission of the first task to the |
1490 |
< |
* pool, so must perform secondary initialization: creating |
1468 |
< |
* workQueue array and setting plock to a valid value. It also |
1490 |
> |
* pool, so must perform secondary initialization. It also |
1491 |
|
* detects first submission by an external thread by looking up |
1492 |
|
* its ThreadLocal, and creates a new shared queue if the one at |
1493 |
< |
* index if empty or contended. The lock bodies must be |
1493 |
> |
* index if empty or contended. The plock lock body must be |
1494 |
|
* exception-free (so no try/finally) so we optimistically |
1495 |
< |
* allocate new queues/arrays outside the locks and throw them |
1496 |
< |
* away if (very rarely) not needed. Note that the plock seq value |
1497 |
< |
* can eventually wrap around zero, but if so harmlessly fails to |
1498 |
< |
* reinitialize. |
1495 |
> |
* allocate new queues outside the lock and throw them away if |
1496 |
> |
* (very rarely) not needed. |
1497 |
> |
* |
1498 |
> |
* Secondary initialization occurs when plock is zero, to create |
1499 |
> |
* workQueue array and set plock to a valid value. This lock body |
1500 |
> |
* must also be exception-free. Because the plock seq value can |
1501 |
> |
* eventually wrap around zero, this method harmlessly fails to |
1502 |
> |
* reinitialize if workQueues exists, while still advancing plock. |
1503 |
|
*/ |
1504 |
|
private void fullExternalPush(ForkJoinTask<?> task) { |
1505 |
< |
for (Submitter z = null;;) { |
1506 |
< |
WorkQueue[] ws; WorkQueue q; int ps, m, r, s; |
1507 |
< |
if ((ps = plock) < 0) |
1505 |
> |
int r = 0; // random index seed |
1506 |
> |
for (Submitter z = submitters.get();;) { |
1507 |
> |
WorkQueue[] ws; WorkQueue q; int ps, m, k; |
1508 |
> |
if (z == null) { |
1509 |
> |
if (U.compareAndSwapInt(this, INDEXSEED, r = indexSeed, |
1510 |
> |
r += SEED_INCREMENT) && r != 0) |
1511 |
> |
submitters.set(z = new Submitter(r)); |
1512 |
> |
} |
1513 |
> |
else if (r == 0) { // move to a different index |
1514 |
> |
r = z.seed; |
1515 |
> |
r ^= r << 13; // same xorshift as WorkQueues |
1516 |
> |
r ^= r >>> 17; |
1517 |
> |
z.seed = r ^ (r << 5); |
1518 |
> |
} |
1519 |
> |
else if ((ps = plock) < 0) |
1520 |
|
throw new RejectedExecutionException(); |
1521 |
< |
else if ((ws = workQueues) == null || (m = ws.length - 1) < 0) { |
1522 |
< |
int n = parallelism - 1; n |= n >>> 1; n |= n >>> 2; |
1523 |
< |
n |= n >>> 4; n |= n >>> 8; n |= n >>> 16; |
1524 |
< |
WorkQueue[] nws = new WorkQueue[(n + 1) << 1]; // power of two |
1525 |
< |
if ((ps & PL_LOCK) != 0 || |
1521 |
> |
else if (ps == 0 || (ws = workQueues) == null || |
1522 |
> |
(m = ws.length - 1) < 0) { // initialize workQueues |
1523 |
> |
int p = config & SMASK; // find power of two table size |
1524 |
> |
int n = (p > 1) ? p - 1 : 1; // ensure at least 2 slots |
1525 |
> |
n |= n >>> 1; n |= n >>> 2; n |= n >>> 4; |
1526 |
> |
n |= n >>> 8; n |= n >>> 16; n = (n + 1) << 1; |
1527 |
> |
WorkQueue[] nws = ((ws = workQueues) == null || ws.length == 0 ? |
1528 |
> |
new WorkQueue[n] : null); |
1529 |
> |
if (((ps = plock) & PL_LOCK) != 0 || |
1530 |
|
!U.compareAndSwapInt(this, PLOCK, ps, ps += PL_LOCK)) |
1531 |
|
ps = acquirePlock(); |
1532 |
< |
if ((ws = workQueues) == null) |
1532 |
> |
if (((ws = workQueues) == null || ws.length == 0) && nws != null) |
1533 |
|
workQueues = nws; |
1534 |
|
int nps = (ps & SHUTDOWN) | ((ps + PL_LOCK) & ~SHUTDOWN); |
1535 |
|
if (!U.compareAndSwapInt(this, PLOCK, ps, nps)) |
1536 |
|
releasePlock(nps); |
1537 |
|
} |
1538 |
< |
else if (z == null && (z = submitters.get()) == null) { |
1539 |
< |
if (U.compareAndSwapInt(this, INDEXSEED, |
1540 |
< |
s = indexSeed, s += SEED_INCREMENT) && |
1541 |
< |
s != 0) // skip 0 |
1542 |
< |
submitters.set(z = new Submitter(s)); |
1538 |
> |
else if ((q = ws[k = r & m & SQMASK]) != null) { |
1539 |
> |
if (q.qlock == 0 && U.compareAndSwapInt(q, QLOCK, 0, 1)) { |
1540 |
> |
ForkJoinTask<?>[] a = q.array; |
1541 |
> |
int s = q.top; |
1542 |
> |
boolean submitted = false; |
1543 |
> |
try { // locked version of push |
1544 |
> |
if ((a != null && a.length > s + 1 - q.base) || |
1545 |
> |
(a = q.growArray()) != null) { // must presize |
1546 |
> |
int j = (((a.length - 1) & s) << ASHIFT) + ABASE; |
1547 |
> |
U.putOrderedObject(a, j, task); |
1548 |
> |
q.top = s + 1; |
1549 |
> |
submitted = true; |
1550 |
> |
} |
1551 |
> |
} finally { |
1552 |
> |
q.qlock = 0; // unlock |
1553 |
> |
} |
1554 |
> |
if (submitted) { |
1555 |
> |
signalWork(q); |
1556 |
> |
return; |
1557 |
> |
} |
1558 |
> |
} |
1559 |
> |
r = 0; // move on failure |
1560 |
|
} |
1561 |
< |
else { |
1562 |
< |
int k = (r = z.seed) & m & SQMASK; |
1563 |
< |
if ((q = ws[k]) == null && (ps & PL_LOCK) == 0) { |
1564 |
< |
(q = new WorkQueue(this, null, SHARED_QUEUE)).poolIndex = k; |
1565 |
< |
if (((ps = plock) & PL_LOCK) != 0 || |
1566 |
< |
!U.compareAndSwapInt(this, PLOCK, ps, ps += PL_LOCK)) |
1567 |
< |
ps = acquirePlock(); |
1568 |
< |
WorkQueue w = null; |
1569 |
< |
if ((ws = workQueues) != null && k < ws.length && |
1570 |
< |
(w = ws[k]) == null) |
1512 |
< |
ws[k] = q; |
1513 |
< |
else |
1514 |
< |
q = w; |
1515 |
< |
int nps = (ps & SHUTDOWN) | ((ps + PL_LOCK) & ~SHUTDOWN); |
1516 |
< |
if (!U.compareAndSwapInt(this, PLOCK, ps, nps)) |
1517 |
< |
releasePlock(nps); |
1518 |
< |
} |
1519 |
< |
if (q != null && q.qlock == 0 && q.fullPush(task, false)) |
1520 |
< |
return; |
1521 |
< |
r ^= r << 13; // same xorshift as WorkQueues |
1522 |
< |
r ^= r >>> 17; |
1523 |
< |
z.seed = r ^= r << 5; // move to a different index |
1561 |
> |
else if (((ps = plock) & PL_LOCK) == 0) { // create new queue |
1562 |
> |
q = new WorkQueue(this, null, SHARED_QUEUE, r); |
1563 |
> |
if (((ps = plock) & PL_LOCK) != 0 || |
1564 |
> |
!U.compareAndSwapInt(this, PLOCK, ps, ps += PL_LOCK)) |
1565 |
> |
ps = acquirePlock(); |
1566 |
> |
if ((ws = workQueues) != null && k < ws.length && ws[k] == null) |
1567 |
> |
ws[k] = q; |
1568 |
> |
int nps = (ps & SHUTDOWN) | ((ps + PL_LOCK) & ~SHUTDOWN); |
1569 |
> |
if (!U.compareAndSwapInt(this, PLOCK, ps, nps)) |
1570 |
> |
releasePlock(nps); |
1571 |
|
} |
1572 |
+ |
else |
1573 |
+ |
r = 0; // try elsewhere while lock held |
1574 |
|
} |
1575 |
|
} |
1576 |
|
|
1585 |
|
} |
1586 |
|
|
1587 |
|
/** |
1588 |
< |
* Tries to create (at most one) or activate (possibly several) |
1540 |
< |
* workers if too few are active. On contention failure, continues |
1541 |
< |
* until at least one worker is signalled or the given queue is |
1542 |
< |
* empty or all workers are active. |
1588 |
> |
* Tries to create or activate a worker if too few are active. |
1589 |
|
* |
1590 |
< |
* @param q if non-null, the queue holding tasks to be signalled |
1545 |
< |
* @param signals the target number of signals. |
1590 |
> |
* @param q the (non-null) queue holding tasks to be signalled |
1591 |
|
*/ |
1592 |
< |
final void signalWork(WorkQueue q, int signals) { |
1593 |
< |
long c; int e, u, i; WorkQueue[] ws; WorkQueue w; Thread p; |
1592 |
> |
final void signalWork(WorkQueue q) { |
1593 |
> |
int hint = q.poolIndex; |
1594 |
> |
long c; int e, u, i, n; WorkQueue[] ws; WorkQueue w; Thread p; |
1595 |
|
while ((u = (int)((c = ctl) >>> 32)) < 0) { |
1596 |
|
if ((e = (int)c) > 0) { |
1597 |
|
if ((ws = workQueues) != null && ws.length > (i = e & SMASK) && |
1599 |
|
long nc = (((long)(w.nextWait & E_MASK)) | |
1600 |
|
((long)(u + UAC_UNIT) << 32)); |
1601 |
|
if (U.compareAndSwapLong(this, CTL, c, nc)) { |
1602 |
+ |
w.hint = hint; |
1603 |
|
w.eventCount = (e + E_SEQ) & E_MASK; |
1604 |
|
if ((p = w.parker) != null) |
1605 |
|
U.unpark(p); |
1606 |
< |
if (--signals <= 0) |
1560 |
< |
break; |
1606 |
> |
break; |
1607 |
|
} |
1608 |
< |
else |
1563 |
< |
signals = 1; |
1564 |
< |
if ((q != null && q.queueSize() == 0)) |
1608 |
> |
if (q.top - q.base <= 0) |
1609 |
|
break; |
1610 |
|
} |
1611 |
|
else |
1612 |
|
break; |
1613 |
|
} |
1614 |
< |
else if (e == 0 && (u & SHORT_SIGN) != 0) { |
1615 |
< |
long nc = (long)(((u + UTC_UNIT) & UTC_MASK) | |
1616 |
< |
((u + UAC_UNIT) & UAC_MASK)) << 32; |
1573 |
< |
if (U.compareAndSwapLong(this, CTL, c, nc)) { |
1574 |
< |
ForkJoinWorkerThread wt = null; |
1575 |
< |
Throwable ex = null; |
1576 |
< |
boolean started = false; |
1577 |
< |
try { |
1578 |
< |
ForkJoinWorkerThreadFactory fac; |
1579 |
< |
if ((fac = factory) != null && |
1580 |
< |
(wt = fac.newThread(this)) != null) { |
1581 |
< |
wt.start(); |
1582 |
< |
started = true; |
1583 |
< |
} |
1584 |
< |
} catch (Throwable rex) { |
1585 |
< |
ex = rex; |
1586 |
< |
} |
1587 |
< |
if (!started) |
1588 |
< |
deregisterWorker(wt, ex); // adjust counts on failure |
1589 |
< |
break; |
1590 |
< |
} |
1591 |
< |
} |
1592 |
< |
else |
1614 |
> |
else { |
1615 |
> |
if ((short)u < 0) |
1616 |
> |
tryAddWorker(); |
1617 |
|
break; |
1618 |
+ |
} |
1619 |
|
} |
1620 |
|
} |
1621 |
|
|
1625 |
|
* Top-level runloop for workers, called by ForkJoinWorkerThread.run. |
1626 |
|
*/ |
1627 |
|
final void runWorker(WorkQueue w) { |
1628 |
< |
// initialize queue array in this thread |
1604 |
< |
w.array = new ForkJoinTask<?>[WorkQueue.INITIAL_QUEUE_CAPACITY]; |
1628 |
> |
w.growArray(); // allocate queue |
1629 |
|
do { w.runTask(scan(w)); } while (w.qlock >= 0); |
1630 |
|
} |
1631 |
|
|
1637 |
|
* contention, or state changes that indicate possible success on |
1638 |
|
* re-invocation. |
1639 |
|
* |
1640 |
< |
* The scan searches for tasks across a random permutation of |
1641 |
< |
* queues (starting at a random index and stepping by a random |
1642 |
< |
* relative prime, checking each at least once). The scan |
1643 |
< |
* terminates upon either finding a non-empty queue, or completing |
1644 |
< |
* the sweep. If the worker is not inactivated, it takes and |
1645 |
< |
* returns a task from this queue. Otherwise, if not activated, it |
1646 |
< |
* signals workers (that may include itself) and returns so caller |
1647 |
< |
* can retry. Also returns for trtry if the worker array may have |
1648 |
< |
* changed during an empty scan. On failure to find a task, we |
1649 |
< |
* take one of the following actions, after which the caller will |
1650 |
< |
* retry calling this method unless terminated. |
1640 |
> |
* The scan searches for tasks across queues (starting at a random |
1641 |
> |
* index, and relying on registerWorker to irregularly scatter |
1642 |
> |
* them within array to avoid bias), checking each at least twice. |
1643 |
> |
* The scan terminates upon either finding a non-empty queue, or |
1644 |
> |
* completing the sweep. If the worker is not inactivated, it |
1645 |
> |
* takes and returns a task from this queue. Otherwise, if not |
1646 |
> |
* activated, it signals workers (that may include itself) and |
1647 |
> |
* returns so caller can retry. Also returns for true if the |
1648 |
> |
* worker array may have changed during an empty scan. On failure |
1649 |
> |
* to find a task, we take one of the following actions, after |
1650 |
> |
* which the caller will retry calling this method unless |
1651 |
> |
* terminated. |
1652 |
|
* |
1653 |
|
* * If pool is terminating, terminate the worker. |
1654 |
|
* |
1655 |
|
* * If not already enqueued, try to inactivate and enqueue the |
1656 |
|
* worker on wait queue. Or, if inactivating has caused the pool |
1657 |
< |
* to be quiescent, relay to idleAwaitWork to check for |
1658 |
< |
* termination and possibly shrink pool. |
1657 |
> |
* to be quiescent, relay to idleAwaitWork to possibly shrink |
1658 |
> |
* pool. |
1659 |
|
* |
1660 |
|
* * If already enqueued and none of the above apply, possibly |
1661 |
< |
* (with 1/2 probability) park awaiting signal, else lingering to |
1662 |
< |
* help scan and signal. |
1661 |
> |
* park awaiting signal, else lingering to help scan and signal. |
1662 |
> |
* |
1663 |
> |
* * If a non-empty queue discovered or left as a hint, |
1664 |
> |
* help wake up other workers before return. |
1665 |
|
* |
1666 |
|
* @param w the worker (via its WorkQueue) |
1667 |
|
* @return a task or null if none found |
1668 |
|
*/ |
1669 |
|
private final ForkJoinTask<?> scan(WorkQueue w) { |
1670 |
< |
WorkQueue[] ws; WorkQueue q; // first update random seed |
1671 |
< |
int r = w.seed; r ^= r << 13; r ^= r >>> 17; w.seed = r ^= r << 5; |
1672 |
< |
int ps = plock, m; // volatile read order matters |
1673 |
< |
if ((ws = workQueues) != null && (m = ws.length - 1) > 0) { |
1674 |
< |
int ec = w.eventCount; // ec is negative if inactive |
1675 |
< |
int step = (r >>> 16) | 1; // relatively prime |
1676 |
< |
for (int j = (m + 1) << 2; ; --j, r += step) { |
1677 |
< |
ForkJoinTask<?> t; ForkJoinTask<?>[] a; int b, n; |
1678 |
< |
if ((q = ws[r & m]) != null && (b = q.base) - q.top < 0 && |
1679 |
< |
(a = q.array) != null) { // probably nonempty |
1670 |
> |
WorkQueue[] ws; int m; |
1671 |
> |
int ps = plock; // read plock before ws |
1672 |
> |
if (w != null && (ws = workQueues) != null && (m = ws.length - 1) >= 0) { |
1673 |
> |
int ec = w.eventCount; // ec is negative if inactive |
1674 |
> |
int r = w.seed; r ^= r << 13; r ^= r >>> 17; w.seed = r ^= r << 5; |
1675 |
> |
w.hint = -1; // update seed and clear hint |
1676 |
> |
int j = ((m + m + 1) | MIN_SCAN) & MAX_SCAN; |
1677 |
> |
do { |
1678 |
> |
WorkQueue q; ForkJoinTask<?>[] a; int b; |
1679 |
> |
if ((q = ws[(r + j) & m]) != null && (b = q.base) - q.top < 0 && |
1680 |
> |
(a = q.array) != null) { // probably nonempty |
1681 |
|
int i = (((a.length - 1) & b) << ASHIFT) + ABASE; |
1682 |
< |
t = (ForkJoinTask<?>)U.getObjectVolatile(a, i); |
1682 |
> |
ForkJoinTask<?> t = (ForkJoinTask<?>) |
1683 |
> |
U.getObjectVolatile(a, i); |
1684 |
|
if (q.base == b && ec >= 0 && t != null && |
1685 |
|
U.compareAndSwapObject(a, i, t, null)) { |
1686 |
< |
if ((n = q.top - (q.base = b + 1)) > 0) |
1687 |
< |
signalWork(q, n); |
1688 |
< |
return t; // taken |
1689 |
< |
} |
1690 |
< |
if (j < m || (ec < 0 && (ec = w.eventCount) < 0)) { |
1691 |
< |
if ((n = q.queueSize() - 1) > 0) |
1692 |
< |
signalWork(q, n); |
1693 |
< |
break; // let caller retry after signal |
1694 |
< |
} |
1695 |
< |
} |
1696 |
< |
else if (j < 0) { // end of scan |
1697 |
< |
long c = ctl; int e; |
1698 |
< |
if (plock != ps) // incomplete sweep |
1699 |
< |
break; |
1700 |
< |
if ((e = (int)c) < 0) // pool is terminating |
1701 |
< |
w.qlock = -1; |
1702 |
< |
else if (ec >= 0) { // try to enqueue/inactivate |
1703 |
< |
long nc = ((long)ec | |
1704 |
< |
((c - AC_UNIT) & (AC_MASK|TC_MASK))); |
1705 |
< |
w.nextWait = e; |
1706 |
< |
w.eventCount = ec | INT_SIGN; // mark as inactive |
1707 |
< |
if (ctl != c || |
1708 |
< |
!U.compareAndSwapLong(this, CTL, c, nc)) |
1709 |
< |
w.eventCount = ec; // unmark on CAS failure |
1710 |
< |
else if ((int)(c >> AC_SHIFT) == 1 - parallelism) |
1711 |
< |
idleAwaitWork(w, nc, c); // quiescent |
1686 |
> |
if ((q.base = b + 1) - q.top < 0) |
1687 |
> |
signalWork(q); |
1688 |
> |
return t; // taken |
1689 |
> |
} |
1690 |
> |
else if ((ec < 0 || j < m) && (int)(ctl >> AC_SHIFT) <= 0) { |
1691 |
> |
w.hint = (r + j) & m; // help signal below |
1692 |
> |
break; // cannot take |
1693 |
> |
} |
1694 |
> |
} |
1695 |
> |
} while (--j >= 0); |
1696 |
> |
|
1697 |
> |
int h, e, ns; long c, sc; WorkQueue q; |
1698 |
> |
if ((ns = w.nsteals) != 0) { |
1699 |
> |
if (U.compareAndSwapLong(this, STEALCOUNT, |
1700 |
> |
sc = stealCount, sc + ns)) |
1701 |
> |
w.nsteals = 0; // collect steals and rescan |
1702 |
> |
} |
1703 |
> |
else if (plock != ps) // consistency check |
1704 |
> |
; // skip |
1705 |
> |
else if ((e = (int)(c = ctl)) < 0) |
1706 |
> |
w.qlock = -1; // pool is terminating |
1707 |
> |
else { |
1708 |
> |
if ((h = w.hint) < 0) { |
1709 |
> |
if (ec >= 0) { // try to enqueue/inactivate |
1710 |
> |
long nc = (((long)ec | |
1711 |
> |
((c - AC_UNIT) & (AC_MASK|TC_MASK)))); |
1712 |
> |
w.nextWait = e; // link and mark inactive |
1713 |
> |
w.eventCount = ec | INT_SIGN; |
1714 |
> |
if (ctl != c || !U.compareAndSwapLong(this, CTL, c, nc)) |
1715 |
> |
w.eventCount = ec; // unmark on CAS failure |
1716 |
> |
else if ((int)(c >> AC_SHIFT) == 1 - (config & SMASK)) |
1717 |
> |
idleAwaitWork(w, nc, c); |
1718 |
|
} |
1719 |
< |
else if (w.seed >= 0 && w.eventCount < 0) { |
1719 |
> |
else if (w.eventCount < 0 && ctl == c) { |
1720 |
|
Thread wt = Thread.currentThread(); |
1721 |
< |
Thread.interrupted(); // clear status |
1721 |
> |
Thread.interrupted(); // clear status |
1722 |
|
U.putObject(wt, PARKBLOCKER, this); |
1723 |
< |
w.parker = wt; // emulate LockSupport.park |
1724 |
< |
if (w.eventCount < 0) // recheck |
1725 |
< |
U.park(false, 0L); |
1723 |
> |
w.parker = wt; // emulate LockSupport.park |
1724 |
> |
if (w.eventCount < 0) // recheck |
1725 |
> |
U.park(false, 0L); // block |
1726 |
|
w.parker = null; |
1727 |
|
U.putObject(wt, PARKBLOCKER, null); |
1728 |
|
} |
1729 |
< |
break; |
1729 |
> |
} |
1730 |
> |
if ((h >= 0 || (h = w.hint) >= 0) && |
1731 |
> |
(ws = workQueues) != null && h < ws.length && |
1732 |
> |
(q = ws[h]) != null) { // signal others before retry |
1733 |
> |
WorkQueue v; Thread p; int u, i, s; |
1734 |
> |
for (int n = (config & SMASK) - 1;;) { |
1735 |
> |
int idleCount = (w.eventCount < 0) ? 0 : -1; |
1736 |
> |
if (((s = idleCount - q.base + q.top) <= n && |
1737 |
> |
(n = s) <= 0) || |
1738 |
> |
(u = (int)((c = ctl) >>> 32)) >= 0 || |
1739 |
> |
(e = (int)c) <= 0 || m < (i = e & SMASK) || |
1740 |
> |
(v = ws[i]) == null) |
1741 |
> |
break; |
1742 |
> |
long nc = (((long)(v.nextWait & E_MASK)) | |
1743 |
> |
((long)(u + UAC_UNIT) << 32)); |
1744 |
> |
if (v.eventCount != (e | INT_SIGN) || |
1745 |
> |
!U.compareAndSwapLong(this, CTL, c, nc)) |
1746 |
> |
break; |
1747 |
> |
v.hint = h; |
1748 |
> |
v.eventCount = (e + E_SEQ) & E_MASK; |
1749 |
> |
if ((p = v.parker) != null) |
1750 |
> |
U.unpark(p); |
1751 |
> |
if (--n <= 0) |
1752 |
> |
break; |
1753 |
> |
} |
1754 |
|
} |
1755 |
|
} |
1756 |
|
} |
1770 |
|
* @param prevCtl the ctl value to restore if thread is terminated |
1771 |
|
*/ |
1772 |
|
private void idleAwaitWork(WorkQueue w, long currentCtl, long prevCtl) { |
1773 |
< |
if (w.eventCount < 0 && |
1774 |
< |
(this == commonPool || !tryTerminate(false, false)) && |
1775 |
< |
(int)prevCtl != 0) { |
1773 |
> |
if (w != null && w.eventCount < 0 && |
1774 |
> |
!tryTerminate(false, false) && (int)prevCtl != 0 && |
1775 |
> |
ctl == currentCtl) { |
1776 |
|
int dc = -(short)(currentCtl >>> TC_SHIFT); |
1777 |
|
long parkTime = dc < 0 ? FAST_IDLE_TIMEOUT: (dc + 1) * IDLE_TIMEOUT; |
1778 |
< |
long deadline = System.nanoTime() + parkTime - 100000L; // 1ms slop |
1778 |
> |
long deadline = System.nanoTime() + parkTime - TIMEOUT_SLOP; |
1779 |
|
Thread wt = Thread.currentThread(); |
1780 |
|
while (ctl == currentCtl) { |
1781 |
|
Thread.interrupted(); // timed variant of version in scan() |
1790 |
|
if (deadline - System.nanoTime() <= 0L && |
1791 |
|
U.compareAndSwapLong(this, CTL, currentCtl, prevCtl)) { |
1792 |
|
w.eventCount = (w.eventCount + E_SEQ) | E_MASK; |
1793 |
+ |
w.hint = -1; |
1794 |
|
w.qlock = -1; // shrink |
1795 |
|
break; |
1796 |
|
} |
1799 |
|
} |
1800 |
|
|
1801 |
|
/** |
1802 |
< |
* Scans through queues looking for work while joining a task; |
1803 |
< |
* if any are present, signals. |
1802 |
> |
* Scans through queues looking for work while joining a task; if |
1803 |
> |
* any present, signals. May return early if more signalling is |
1804 |
> |
* detectably unneeded. |
1805 |
|
* |
1806 |
< |
* @param task to return early if done |
1806 |
> |
* @param task return early if done |
1807 |
|
* @param origin an index to start scan |
1808 |
|
*/ |
1809 |
< |
final int helpSignal(ForkJoinTask<?> task, int origin) { |
1810 |
< |
WorkQueue[] ws; WorkQueue q; int m, n, s; |
1811 |
< |
if (task != null && (ws = workQueues) != null && |
1812 |
< |
(m = ws.length - 1) >= 0) { |
1813 |
< |
for (int i = 0; i <= m; ++i) { |
1814 |
< |
if ((s = task.status) < 0) |
1815 |
< |
return s; |
1816 |
< |
if ((q = ws[(i + origin) & m]) != null && |
1817 |
< |
(n = q.queueSize()) > 0) { |
1818 |
< |
signalWork(q, n); |
1819 |
< |
if ((int)(ctl >> AC_SHIFT) >= 0) |
1809 |
> |
private void helpSignal(ForkJoinTask<?> task, int origin) { |
1810 |
> |
WorkQueue[] ws; WorkQueue w; Thread p; long c; int m, u, e, i, s; |
1811 |
> |
if (task != null && task.status >= 0 && |
1812 |
> |
(u = (int)(ctl >>> 32)) < 0 && (u >> UAC_SHIFT) < 0 && |
1813 |
> |
(ws = workQueues) != null && (m = ws.length - 1) >= 0) { |
1814 |
> |
outer: for (int k = origin, j = m; j >= 0; --j) { |
1815 |
> |
WorkQueue q = ws[k++ & m]; |
1816 |
> |
for (int n = m;;) { // limit to at most m signals |
1817 |
> |
if (task.status < 0) |
1818 |
> |
break outer; |
1819 |
> |
if (q == null || |
1820 |
> |
((s = -q.base + q.top) <= n && (n = s) <= 0)) |
1821 |
|
break; |
1822 |
+ |
if ((u = (int)((c = ctl) >>> 32)) >= 0 || |
1823 |
+ |
(e = (int)c) <= 0 || m < (i = e & SMASK) || |
1824 |
+ |
(w = ws[i]) == null) |
1825 |
+ |
break outer; |
1826 |
+ |
long nc = (((long)(w.nextWait & E_MASK)) | |
1827 |
+ |
((long)(u + UAC_UNIT) << 32)); |
1828 |
+ |
if (w.eventCount != (e | INT_SIGN)) |
1829 |
+ |
break outer; |
1830 |
+ |
if (U.compareAndSwapLong(this, CTL, c, nc)) { |
1831 |
+ |
w.eventCount = (e + E_SEQ) & E_MASK; |
1832 |
+ |
if ((p = w.parker) != null) |
1833 |
+ |
U.unpark(p); |
1834 |
+ |
if (--n <= 0) |
1835 |
+ |
break; |
1836 |
+ |
} |
1837 |
|
} |
1838 |
|
} |
1839 |
|
} |
1763 |
– |
return 0; |
1840 |
|
} |
1841 |
|
|
1842 |
|
/** |
1870 |
|
} |
1871 |
|
if ((ws = workQueues) == null || (m = ws.length - 1) <= 0) |
1872 |
|
break restart; // shutting down |
1873 |
< |
if ((v = ws[h = (j.stealHint | 1) & m]) == null || |
1873 |
> |
if ((v = ws[h = (j.hint | 1) & m]) == null || |
1874 |
|
v.currentSteal != subtask) { |
1875 |
|
for (int origin = h;;) { // find stealer |
1876 |
|
if (((h = (h + 2) & m) & 15) == 1 && |
1878 |
|
continue restart; // occasional staleness check |
1879 |
|
if ((v = ws[h]) != null && |
1880 |
|
v.currentSteal == subtask) { |
1881 |
< |
j.stealHint = h; // save hint |
1881 |
> |
j.hint = h; // save hint |
1882 |
|
break; |
1883 |
|
} |
1884 |
|
if (h == origin) |
1927 |
|
|
1928 |
|
/** |
1929 |
|
* Analog of tryHelpStealer for CountedCompleters. Tries to steal |
1930 |
< |
* and run tasks within the target's computation |
1930 |
> |
* and run tasks within the target's computation. |
1931 |
|
* |
1932 |
|
* @param task the task to join |
1933 |
|
* @param mode if shared, exit upon completing any task |
1934 |
|
* if all workers are active |
1859 |
– |
* |
1935 |
|
*/ |
1936 |
|
private int helpComplete(ForkJoinTask<?> task, int mode) { |
1937 |
< |
WorkQueue[] ws; WorkQueue q; int m, n, s; |
1937 |
> |
WorkQueue[] ws; WorkQueue q; int m, n, s, u; |
1938 |
|
if (task != null && (ws = workQueues) != null && |
1939 |
|
(m = ws.length - 1) >= 0) { |
1940 |
|
for (int j = 1, origin = j;;) { |
1942 |
|
return s; |
1943 |
|
if ((q = ws[j & m]) != null && q.pollAndExecCC(task)) { |
1944 |
|
origin = j; |
1945 |
< |
if (mode == SHARED_QUEUE && (int)(ctl >> AC_SHIFT) >= 0) |
1945 |
> |
if (mode == SHARED_QUEUE && |
1946 |
> |
((u = (int)(ctl >>> 32)) >= 0 || (u >> UAC_SHIFT) >= 0)) |
1947 |
|
break; |
1948 |
|
} |
1949 |
|
else if ((j = (j + 2) & m) == origin) |
1961 |
|
* may become starved. |
1962 |
|
*/ |
1963 |
|
final boolean tryCompensate() { |
1964 |
< |
int pc = parallelism, e, u, i, tc; long c; |
1964 |
> |
int pc = config & SMASK, e, i, tc; long c; |
1965 |
|
WorkQueue[] ws; WorkQueue w; Thread p; |
1966 |
< |
if ((e = (int)(c = ctl)) >= 0 && (ws = workQueues) != null) { |
1966 |
> |
if ((ws = workQueues) != null && (e = (int)(c = ctl)) >= 0) { |
1967 |
|
if (e != 0 && (i = e & SMASK) < ws.length && |
1968 |
|
(w = ws[i]) != null && w.eventCount == (e | INT_SIGN)) { |
1969 |
|
long nc = ((long)(w.nextWait & E_MASK) | |
1975 |
|
return true; // replace with idle worker |
1976 |
|
} |
1977 |
|
} |
1978 |
< |
else if ((short)((u = (int)(c >>> 32)) >>> UTC_SHIFT) >= 0 && |
1979 |
< |
(u >> UAC_SHIFT) + pc > 1) { |
1978 |
> |
else if ((tc = (short)(c >>> TC_SHIFT)) >= 0 && |
1979 |
> |
(int)(c >> AC_SHIFT) + pc > 1) { |
1980 |
|
long nc = ((c - AC_UNIT) & AC_MASK) | (c & ~AC_MASK); |
1981 |
|
if (U.compareAndSwapLong(this, CTL, c, nc)) |
1982 |
< |
return true; // no compensation |
1982 |
> |
return true; // no compensation |
1983 |
|
} |
1984 |
< |
else if ((tc = u + pc) < MAX_CAP) { |
1984 |
> |
else if (tc + pc < MAX_CAP) { |
1985 |
|
long nc = ((c + TC_UNIT) & TC_MASK) | (c & ~TC_MASK); |
1986 |
|
if (U.compareAndSwapLong(this, CTL, c, nc)) { |
1987 |
+ |
ForkJoinWorkerThreadFactory fac; |
1988 |
|
Throwable ex = null; |
1989 |
|
ForkJoinWorkerThread wt = null; |
1990 |
|
try { |
1914 |
– |
ForkJoinWorkerThreadFactory fac; |
1991 |
|
if ((fac = factory) != null && |
1992 |
|
(wt = fac.newThread(this)) != null) { |
1993 |
|
wt.start(); |
1996 |
|
} catch (Throwable rex) { |
1997 |
|
ex = rex; |
1998 |
|
} |
1999 |
< |
deregisterWorker(wt, ex); // adjust counts etc |
1999 |
> |
deregisterWorker(wt, ex); // clean up and return false |
2000 |
|
} |
2001 |
|
} |
2002 |
|
} |
2015 |
|
if (joiner != null && task != null && (s = task.status) >= 0) { |
2016 |
|
ForkJoinTask<?> prevJoin = joiner.currentJoin; |
2017 |
|
joiner.currentJoin = task; |
2018 |
< |
do {} while ((s = task.status) >= 0 && |
1943 |
< |
joiner.queueSize() > 0 && |
2018 |
> |
do {} while ((s = task.status) >= 0 && !joiner.isEmpty() && |
2019 |
|
joiner.tryRemoveAndExec(task)); // process local tasks |
2020 |
< |
if (s >= 0 && (s = task.status) >= 0 && |
2021 |
< |
(s = helpSignal(task, joiner.poolIndex)) >= 0 && |
2022 |
< |
(task instanceof CountedCompleter)) |
2023 |
< |
s = helpComplete(task, LIFO_QUEUE); |
2020 |
> |
if (s >= 0 && (s = task.status) >= 0) { |
2021 |
> |
helpSignal(task, joiner.poolIndex); |
2022 |
> |
if ((s = task.status) >= 0 && |
2023 |
> |
(task instanceof CountedCompleter)) |
2024 |
> |
s = helpComplete(task, LIFO_QUEUE); |
2025 |
> |
} |
2026 |
|
while (s >= 0 && (s = task.status) >= 0) { |
2027 |
< |
if ((joiner.queueSize() > 0 || // try helping |
2027 |
> |
if ((!joiner.isEmpty() || // try helping |
2028 |
|
(s = tryHelpStealer(joiner, task)) == 0) && |
2029 |
< |
(s = task.status) >= 0 && tryCompensate()) { |
2030 |
< |
if (task.trySetSignal() && (s = task.status) >= 0) { |
2031 |
< |
synchronized (task) { |
2032 |
< |
if (task.status >= 0) { |
2033 |
< |
try { // see ForkJoinTask |
2034 |
< |
task.wait(); // for explanation |
2035 |
< |
} catch (InterruptedException ie) { |
2029 |
> |
(s = task.status) >= 0) { |
2030 |
> |
helpSignal(task, joiner.poolIndex); |
2031 |
> |
if ((s = task.status) >= 0 && tryCompensate()) { |
2032 |
> |
if (task.trySetSignal() && (s = task.status) >= 0) { |
2033 |
> |
synchronized (task) { |
2034 |
> |
if (task.status >= 0) { |
2035 |
> |
try { // see ForkJoinTask |
2036 |
> |
task.wait(); // for explanation |
2037 |
> |
} catch (InterruptedException ie) { |
2038 |
> |
} |
2039 |
|
} |
2040 |
+ |
else |
2041 |
+ |
task.notifyAll(); |
2042 |
|
} |
1961 |
– |
else |
1962 |
– |
task.notifyAll(); |
2043 |
|
} |
2044 |
+ |
long c; // re-activate |
2045 |
+ |
do {} while (!U.compareAndSwapLong |
2046 |
+ |
(this, CTL, c = ctl, c + AC_UNIT)); |
2047 |
|
} |
1965 |
– |
long c; // re-activate |
1966 |
– |
do {} while (!U.compareAndSwapLong |
1967 |
– |
(this, CTL, c = ctl, c + AC_UNIT)); |
2048 |
|
} |
2049 |
|
} |
2050 |
|
joiner.currentJoin = prevJoin; |
2065 |
|
if (joiner != null && task != null && (s = task.status) >= 0) { |
2066 |
|
ForkJoinTask<?> prevJoin = joiner.currentJoin; |
2067 |
|
joiner.currentJoin = task; |
2068 |
< |
do {} while ((s = task.status) >= 0 && |
1989 |
< |
joiner.queueSize() > 0 && |
2068 |
> |
do {} while ((s = task.status) >= 0 && !joiner.isEmpty() && |
2069 |
|
joiner.tryRemoveAndExec(task)); |
2070 |
< |
if (s >= 0 && (s = task.status) >= 0 && |
2071 |
< |
(s = helpSignal(task, joiner.poolIndex)) >= 0 && |
2072 |
< |
(task instanceof CountedCompleter)) |
2073 |
< |
s = helpComplete(task, LIFO_QUEUE); |
2074 |
< |
if (s >= 0 && joiner.queueSize() == 0) { |
2070 |
> |
if (s >= 0 && (s = task.status) >= 0) { |
2071 |
> |
helpSignal(task, joiner.poolIndex); |
2072 |
> |
if ((s = task.status) >= 0 && |
2073 |
> |
(task instanceof CountedCompleter)) |
2074 |
> |
s = helpComplete(task, LIFO_QUEUE); |
2075 |
> |
} |
2076 |
> |
if (s >= 0 && joiner.isEmpty()) { |
2077 |
|
do {} while (task.status >= 0 && |
2078 |
|
tryHelpStealer(joiner, task) > 0); |
2079 |
|
} |
2083 |
|
|
2084 |
|
/** |
2085 |
|
* Returns a (probably) non-empty steal queue, if one is found |
2086 |
< |
* during a random, then cyclic scan, else null. This method must |
2087 |
< |
* be retried by caller if, by the time it tries to use the queue, |
2007 |
< |
* it is empty. |
2086 |
> |
* during a scan, else null. This method must be retried by |
2087 |
> |
* caller if, by the time it tries to use the queue, it is empty. |
2088 |
|
* @param r a (random) seed for scanning |
2089 |
|
*/ |
2090 |
|
private WorkQueue findNonEmptyStealQueue(int r) { |
2091 |
< |
int step = (r >>> 16) | 1; |
2092 |
< |
for (WorkQueue[] ws;;) { |
2093 |
< |
int ps = plock, m; |
2094 |
< |
if ((ws = workQueues) == null || (m = ws.length - 1) < 1) |
2095 |
< |
return null; |
2096 |
< |
for (int j = (m + 1) << 2; ; r += step) { |
2097 |
< |
WorkQueue q = ws[((r << 1) | 1) & m]; |
2018 |
< |
if (q != null && q.queueSize() > 0) |
2019 |
< |
return q; |
2020 |
< |
else if (--j < 0) { |
2021 |
< |
if (plock == ps) |
2022 |
< |
return null; |
2023 |
< |
break; |
2091 |
> |
for (;;) { |
2092 |
> |
int ps = plock, m; WorkQueue[] ws; WorkQueue q; |
2093 |
> |
if ((ws = workQueues) != null && (m = ws.length - 1) >= 0) { |
2094 |
> |
for (int j = (m + 1) << 2; j >= 0; --j) { |
2095 |
> |
if ((q = ws[(((r + j) << 1) | 1) & m]) != null && |
2096 |
> |
q.base - q.top < 0) |
2097 |
> |
return q; |
2098 |
|
} |
2099 |
|
} |
2100 |
+ |
if (plock == ps) |
2101 |
+ |
return null; |
2102 |
|
} |
2103 |
|
} |
2104 |
|
|
2110 |
|
*/ |
2111 |
|
final void helpQuiescePool(WorkQueue w) { |
2112 |
|
for (boolean active = true;;) { |
2113 |
< |
ForkJoinTask<?> localTask; // exhaust local queue |
2114 |
< |
while ((localTask = w.nextLocalTask()) != null) |
2115 |
< |
localTask.doExec(); |
2116 |
< |
// Similar to loop in scan(), but ignoring submissions |
2117 |
< |
WorkQueue q = findNonEmptyStealQueue(w.nextSeed()); |
2118 |
< |
if (q != null) { |
2119 |
< |
ForkJoinTask<?> t; int b; |
2113 |
> |
long c; WorkQueue q; ForkJoinTask<?> t; int b; |
2114 |
> |
while ((t = w.nextLocalTask()) != null) { |
2115 |
> |
if (w.base - w.top < 0) |
2116 |
> |
signalWork(w); |
2117 |
> |
t.doExec(); |
2118 |
> |
} |
2119 |
> |
if ((q = findNonEmptyStealQueue(w.nextSeed())) != null) { |
2120 |
|
if (!active) { // re-establish active count |
2045 |
– |
long c; |
2121 |
|
active = true; |
2122 |
|
do {} while (!U.compareAndSwapLong |
2123 |
|
(this, CTL, c = ctl, c + AC_UNIT)); |
2124 |
|
} |
2125 |
< |
if ((b = q.base) - q.top < 0 && (t = q.pollAt(b)) != null) |
2125 |
> |
if ((b = q.base) - q.top < 0 && (t = q.pollAt(b)) != null) { |
2126 |
> |
if (q.base - q.top < 0) |
2127 |
> |
signalWork(q); |
2128 |
|
w.runSubtask(t); |
2129 |
+ |
} |
2130 |
|
} |
2131 |
< |
else { |
2132 |
< |
long c; |
2133 |
< |
if (active) { // decrement active count without queuing |
2131 |
> |
else if (active) { // decrement active count without queuing |
2132 |
> |
long nc = (c = ctl) - AC_UNIT; |
2133 |
> |
if ((int)(nc >> AC_SHIFT) + (config & SMASK) == 0) |
2134 |
> |
return; // bypass decrement-then-increment |
2135 |
> |
if (U.compareAndSwapLong(this, CTL, c, nc)) |
2136 |
|
active = false; |
2057 |
– |
do {} while (!U.compareAndSwapLong |
2058 |
– |
(this, CTL, c = ctl, c -= AC_UNIT)); |
2059 |
– |
} |
2060 |
– |
else |
2061 |
– |
c = ctl; // re-increment on exit |
2062 |
– |
if ((int)(c >> AC_SHIFT) + parallelism == 0) { |
2063 |
– |
do {} while (!U.compareAndSwapLong |
2064 |
– |
(this, CTL, c = ctl, c + AC_UNIT)); |
2065 |
– |
break; |
2066 |
– |
} |
2137 |
|
} |
2138 |
+ |
else if ((int)((c = ctl) >> AC_SHIFT) + (config & SMASK) == 0 && |
2139 |
+ |
U.compareAndSwapLong(this, CTL, c, c + AC_UNIT)) |
2140 |
+ |
return; |
2141 |
|
} |
2142 |
|
} |
2143 |
|
|
2153 |
|
return t; |
2154 |
|
if ((q = findNonEmptyStealQueue(w.nextSeed())) == null) |
2155 |
|
return null; |
2156 |
< |
if ((b = q.base) - q.top < 0 && (t = q.pollAt(b)) != null) |
2156 |
> |
if ((b = q.base) - q.top < 0 && (t = q.pollAt(b)) != null) { |
2157 |
> |
if (q.base - q.top < 0) |
2158 |
> |
signalWork(q); |
2159 |
|
return t; |
2160 |
+ |
} |
2161 |
|
} |
2162 |
|
} |
2163 |
|
|
2210 |
|
static int getSurplusQueuedTaskCount() { |
2211 |
|
Thread t; ForkJoinWorkerThread wt; ForkJoinPool pool; WorkQueue q; |
2212 |
|
if (((t = Thread.currentThread()) instanceof ForkJoinWorkerThread)) { |
2213 |
< |
int b = (q = (wt = (ForkJoinWorkerThread)t).workQueue).base; |
2214 |
< |
int p = (pool = wt.pool).parallelism; |
2213 |
> |
int p = (pool = (wt = (ForkJoinWorkerThread)t).pool).config & SMASK; |
2214 |
> |
int n = (q = wt.workQueue).top - q.base; |
2215 |
|
int a = (int)(pool.ctl >> AC_SHIFT) + p; |
2216 |
< |
return q.top - b - (a > (p >>>= 1) ? 0 : |
2217 |
< |
a > (p >>>= 1) ? 1 : |
2218 |
< |
a > (p >>>= 1) ? 2 : |
2219 |
< |
a > (p >>>= 1) ? 4 : |
2220 |
< |
8); |
2216 |
> |
return n - (a > (p >>>= 1) ? 0 : |
2217 |
> |
a > (p >>>= 1) ? 1 : |
2218 |
> |
a > (p >>>= 1) ? 2 : |
2219 |
> |
a > (p >>>= 1) ? 4 : |
2220 |
> |
8); |
2221 |
|
} |
2222 |
|
return 0; |
2223 |
|
} |
2239 |
|
* @return true if now terminating or terminated |
2240 |
|
*/ |
2241 |
|
private boolean tryTerminate(boolean now, boolean enable) { |
2242 |
< |
if (this == commonPool) // cannot shut down |
2242 |
> |
int ps; |
2243 |
> |
if (this == common) // cannot shut down |
2244 |
|
return false; |
2245 |
+ |
if ((ps = plock) >= 0) { // enable by setting plock |
2246 |
+ |
if (!enable) |
2247 |
+ |
return false; |
2248 |
+ |
if ((ps & PL_LOCK) != 0 || |
2249 |
+ |
!U.compareAndSwapInt(this, PLOCK, ps, ps += PL_LOCK)) |
2250 |
+ |
ps = acquirePlock(); |
2251 |
+ |
int nps = ((ps + PL_LOCK) & ~SHUTDOWN) | SHUTDOWN; |
2252 |
+ |
if (!U.compareAndSwapInt(this, PLOCK, ps, nps)) |
2253 |
+ |
releasePlock(nps); |
2254 |
+ |
} |
2255 |
|
for (long c;;) { |
2256 |
< |
if (((c = ctl) & STOP_BIT) != 0) { // already terminating |
2257 |
< |
if ((short)(c >>> TC_SHIFT) == -parallelism) { |
2256 |
> |
if (((c = ctl) & STOP_BIT) != 0) { // already terminating |
2257 |
> |
if ((short)(c >>> TC_SHIFT) == -(config & SMASK)) { |
2258 |
|
synchronized (this) { |
2259 |
< |
notifyAll(); // signal when 0 workers |
2259 |
> |
notifyAll(); // signal when 0 workers |
2260 |
|
} |
2261 |
|
} |
2262 |
|
return true; |
2263 |
|
} |
2264 |
< |
if (plock >= 0) { // not yet enabled |
2265 |
< |
int ps; |
2266 |
< |
if (!enable) |
2264 |
> |
if (!now) { // check if idle & no tasks |
2265 |
> |
WorkQueue[] ws; WorkQueue w; |
2266 |
> |
if ((int)(c >> AC_SHIFT) != -(config & SMASK)) |
2267 |
|
return false; |
2268 |
< |
if (((ps = plock) & PL_LOCK) != 0 || |
2269 |
< |
!U.compareAndSwapInt(this, PLOCK, ps, ps += PL_LOCK)) |
2270 |
< |
ps = acquirePlock(); |
2271 |
< |
int nps = SHUTDOWN; |
2272 |
< |
if (!U.compareAndSwapInt(this, PLOCK, ps, nps)) |
2273 |
< |
releasePlock(nps); |
2274 |
< |
} |
2275 |
< |
if (!now) { // check if idle & no tasks |
2276 |
< |
if ((int)(c >> AC_SHIFT) != -parallelism || |
2277 |
< |
hasQueuedSubmissions()) |
2191 |
< |
return false; |
2192 |
< |
// Check for unqueued inactive workers. One pass suffices. |
2193 |
< |
WorkQueue[] ws = workQueues; WorkQueue w; |
2194 |
< |
if (ws != null) { |
2195 |
< |
for (int i = 1; i < ws.length; i += 2) { |
2196 |
< |
if ((w = ws[i]) != null && w.eventCount >= 0) |
2197 |
< |
return false; |
2268 |
> |
if ((ws = workQueues) != null) { |
2269 |
> |
for (int i = 0; i < ws.length; ++i) { |
2270 |
> |
if ((w = ws[i]) != null) { |
2271 |
> |
if (!w.isEmpty()) { // signal unprocessed tasks |
2272 |
> |
signalWork(w); |
2273 |
> |
return false; |
2274 |
> |
} |
2275 |
> |
if ((i & 1) != 0 && w.eventCount >= 0) |
2276 |
> |
return false; // unqueued inactive worker |
2277 |
> |
} |
2278 |
|
} |
2279 |
|
} |
2280 |
|
} |
2281 |
|
if (U.compareAndSwapLong(this, CTL, c, c | STOP_BIT)) { |
2282 |
|
for (int pass = 0; pass < 3; ++pass) { |
2283 |
< |
WorkQueue[] ws = workQueues; |
2284 |
< |
if (ws != null) { |
2205 |
< |
WorkQueue w; |
2283 |
> |
WorkQueue[] ws; WorkQueue w; Thread wt; |
2284 |
> |
if ((ws = workQueues) != null) { |
2285 |
|
int n = ws.length; |
2286 |
|
for (int i = 0; i < n; ++i) { |
2287 |
|
if ((w = ws[i]) != null) { |
2288 |
|
w.qlock = -1; |
2289 |
|
if (pass > 0) { |
2290 |
|
w.cancelAll(); |
2291 |
< |
if (pass > 1) |
2292 |
< |
w.interruptOwner(); |
2291 |
> |
if (pass > 1 && (wt = w.owner) != null) { |
2292 |
> |
if (!wt.isInterrupted()) { |
2293 |
> |
try { |
2294 |
> |
wt.interrupt(); |
2295 |
> |
} catch (Throwable ignore) { |
2296 |
> |
} |
2297 |
> |
} |
2298 |
> |
U.unpark(wt); |
2299 |
> |
} |
2300 |
|
} |
2301 |
|
} |
2302 |
|
} |
2303 |
|
// Wake up workers parked on event queue |
2304 |
|
int i, e; long cc; Thread p; |
2305 |
|
while ((e = (int)(cc = ctl) & E_MASK) != 0 && |
2306 |
< |
(i = e & SMASK) < n && |
2306 |
> |
(i = e & SMASK) < n && i >= 0 && |
2307 |
|
(w = ws[i]) != null) { |
2308 |
|
long nc = ((long)(w.nextWait & E_MASK) | |
2309 |
|
((cc + AC_UNIT) & AC_MASK) | |
2331 |
|
static WorkQueue commonSubmitterQueue() { |
2332 |
|
ForkJoinPool p; WorkQueue[] ws; int m; Submitter z; |
2333 |
|
return ((z = submitters.get()) != null && |
2334 |
< |
(p = commonPool) != null && |
2334 |
> |
(p = common) != null && |
2335 |
|
(ws = p.workQueues) != null && |
2336 |
|
(m = ws.length - 1) >= 0) ? |
2337 |
|
ws[m & z.seed & SQMASK] : null; |
2342 |
|
*/ |
2343 |
|
static boolean tryExternalUnpush(ForkJoinTask<?> t) { |
2344 |
|
ForkJoinPool p; WorkQueue[] ws; WorkQueue q; Submitter z; |
2345 |
< |
ForkJoinTask<?>[] a; int m, s; long j; |
2346 |
< |
if ((z = submitters.get()) != null && |
2347 |
< |
(p = commonPool) != null && |
2345 |
> |
ForkJoinTask<?>[] a; int m, s; |
2346 |
> |
if (t != null && |
2347 |
> |
(z = submitters.get()) != null && |
2348 |
> |
(p = common) != null && |
2349 |
|
(ws = p.workQueues) != null && |
2350 |
|
(m = ws.length - 1) >= 0 && |
2351 |
|
(q = ws[m & z.seed & SQMASK]) != null && |
2352 |
|
(s = q.top) != q.base && |
2353 |
< |
(a = q.array) != null && |
2354 |
< |
U.getObjectVolatile |
2355 |
< |
(a, j = (((a.length - 1) & (s - 1)) << ASHIFT) + ABASE) == t && |
2356 |
< |
U.compareAndSwapInt(q, QLOCK, 0, 1)) { |
2357 |
< |
if (q.array == a && q.top == s && // recheck |
2358 |
< |
U.compareAndSwapObject(a, j, t, null)) { |
2359 |
< |
q.top = s - 1; |
2353 |
> |
(a = q.array) != null) { |
2354 |
> |
long j = (((a.length - 1) & (s - 1)) << ASHIFT) + ABASE; |
2355 |
> |
if (U.getObject(a, j) == t && |
2356 |
> |
U.compareAndSwapInt(q, QLOCK, 0, 1)) { |
2357 |
> |
if (q.array == a && q.top == s && // recheck |
2358 |
> |
U.compareAndSwapObject(a, j, t, null)) { |
2359 |
> |
q.top = s - 1; |
2360 |
> |
q.qlock = 0; |
2361 |
> |
return true; |
2362 |
> |
} |
2363 |
|
q.qlock = 0; |
2274 |
– |
return true; |
2364 |
|
} |
2276 |
– |
q.qlock = 0; |
2365 |
|
} |
2366 |
|
return false; |
2367 |
|
} |
2376 |
|
if (q != null && (a = q.array) != null && (m = (a.length - 1)) >= 0 && |
2377 |
|
root != null && root.status >= 0) { |
2378 |
|
for (;;) { |
2379 |
< |
int s; Object o; CountedCompleter<?> task = null; |
2379 |
> |
int s, u; Object o; CountedCompleter<?> task = null; |
2380 |
|
if ((s = q.top) - q.base > 0) { |
2381 |
|
long j = ((m & (s - 1)) << ASHIFT) + ABASE; |
2382 |
|
if ((o = U.getObject(a, j)) != null && |
2399 |
|
} |
2400 |
|
if (task != null) |
2401 |
|
task.doExec(); |
2402 |
< |
if (root.status < 0 || (int)(ctl >> AC_SHIFT) >= 0) |
2402 |
> |
if (root.status < 0 || |
2403 |
> |
(u = (int)(ctl >>> 32)) >= 0 || (u >> UAC_SHIFT) >= 0) |
2404 |
|
break; |
2405 |
|
if (task == null) { |
2406 |
< |
if (helpSignal(root, q.poolIndex) >= 0) |
2406 |
> |
helpSignal(root, q.poolIndex); |
2407 |
> |
if (root.status >= 0) |
2408 |
|
helpComplete(root, SHARED_QUEUE); |
2409 |
|
break; |
2410 |
|
} |
2419 |
|
static void externalHelpJoin(ForkJoinTask<?> t) { |
2420 |
|
// Some hard-to-avoid overlap with tryExternalUnpush |
2421 |
|
ForkJoinPool p; WorkQueue[] ws; WorkQueue q, w; Submitter z; |
2422 |
< |
ForkJoinTask<?>[] a; int m, s, n; long j; |
2423 |
< |
if (t != null && t.status >= 0 && |
2422 |
> |
ForkJoinTask<?>[] a; int m, s, n; |
2423 |
> |
if (t != null && |
2424 |
|
(z = submitters.get()) != null && |
2425 |
< |
(p = commonPool) != null && |
2425 |
> |
(p = common) != null && |
2426 |
|
(ws = p.workQueues) != null && |
2427 |
|
(m = ws.length - 1) >= 0 && |
2428 |
|
(q = ws[m & z.seed & SQMASK]) != null && |
2429 |
|
(a = q.array) != null) { |
2430 |
< |
if ((s = q.top) != q.base && |
2431 |
< |
U.getObjectVolatile |
2432 |
< |
(a, j = (((a.length - 1) & (s - 1)) << ASHIFT) + ABASE) == t && |
2433 |
< |
U.compareAndSwapInt(q, QLOCK, 0, 1)) { |
2434 |
< |
if (q.array == a && q.top == s && |
2435 |
< |
U.compareAndSwapObject(a, j, t, null)) { |
2436 |
< |
q.top = s - 1; |
2437 |
< |
q.qlock = 0; |
2438 |
< |
t.doExec(); |
2430 |
> |
int am = a.length - 1; |
2431 |
> |
if ((s = q.top) != q.base) { |
2432 |
> |
long j = ((am & (s - 1)) << ASHIFT) + ABASE; |
2433 |
> |
if (U.getObject(a, j) == t && |
2434 |
> |
U.compareAndSwapInt(q, QLOCK, 0, 1)) { |
2435 |
> |
if (q.array == a && q.top == s && |
2436 |
> |
U.compareAndSwapObject(a, j, t, null)) { |
2437 |
> |
q.top = s - 1; |
2438 |
> |
q.qlock = 0; |
2439 |
> |
t.doExec(); |
2440 |
> |
} |
2441 |
> |
else |
2442 |
> |
q.qlock = 0; |
2443 |
|
} |
2350 |
– |
else |
2351 |
– |
q.qlock = 0; |
2444 |
|
} |
2445 |
|
if (t.status >= 0) { |
2446 |
|
if (t instanceof CountedCompleter) |
2451 |
|
} |
2452 |
|
} |
2453 |
|
|
2362 |
– |
/** |
2363 |
– |
* Restricted version of helpQuiescePool for external callers |
2364 |
– |
*/ |
2365 |
– |
static void externalHelpQuiescePool() { |
2366 |
– |
ForkJoinPool p; ForkJoinTask<?> t; WorkQueue q; int b; |
2367 |
– |
int r = ThreadLocalRandom.current().nextInt(); |
2368 |
– |
if ((p = commonPool) != null && |
2369 |
– |
(q = p.findNonEmptyStealQueue(r)) != null && |
2370 |
– |
(b = q.base) - q.top < 0 && |
2371 |
– |
(t = q.pollAt(b)) != null) |
2372 |
– |
t.doExec(); |
2373 |
– |
} |
2374 |
– |
|
2454 |
|
// Exported methods |
2455 |
|
|
2456 |
|
// Constructors |
2522 |
|
throw new NullPointerException(); |
2523 |
|
if (parallelism <= 0 || parallelism > MAX_CAP) |
2524 |
|
throw new IllegalArgumentException(); |
2446 |
– |
this.parallelism = parallelism; |
2525 |
|
this.factory = factory; |
2526 |
|
this.ueh = handler; |
2527 |
< |
this.localMode = asyncMode ? FIFO_QUEUE : LIFO_QUEUE; |
2527 |
> |
this.config = parallelism | (asyncMode ? (FIFO_QUEUE << 16) : 0); |
2528 |
|
long np = (long)(-parallelism); // offset ctl counts |
2529 |
|
this.ctl = ((np << AC_SHIFT) & AC_MASK) | ((np << TC_SHIFT) & TC_MASK); |
2530 |
|
int pn = nextPoolId(); |
2541 |
|
ForkJoinPool(int parallelism, long ctl, |
2542 |
|
ForkJoinWorkerThreadFactory factory, |
2543 |
|
Thread.UncaughtExceptionHandler handler) { |
2544 |
< |
this.parallelism = parallelism; |
2544 |
> |
this.config = parallelism; |
2545 |
|
this.ctl = ctl; |
2546 |
|
this.factory = factory; |
2547 |
|
this.ueh = handler; |
2470 |
– |
this.localMode = LIFO_QUEUE; |
2548 |
|
this.workerNamePrefix = "ForkJoinPool.commonPool-worker-"; |
2549 |
|
} |
2550 |
|
|
2551 |
|
/** |
2552 |
< |
* Returns the common pool instance. |
2552 |
> |
* Returns the common pool instance. This pool is statically |
2553 |
> |
* constructed; its run state is unaffected by attempts to {@link |
2554 |
> |
* #shutdown} or {@link #shutdownNow}. However this pool and any |
2555 |
> |
* ongoing processing are automatically terminated upon program |
2556 |
> |
* {@link System#exit}. Any program that relies on asynchronous |
2557 |
> |
* task processing to complete before program termination should |
2558 |
> |
* invoke {@link #quiesceCommonPool}, or the timeout-based {@code |
2559 |
> |
* commonPool().}{@link #awaitQuiescence}, before exit. |
2560 |
|
* |
2561 |
|
* @return the common pool instance |
2562 |
|
*/ |
2563 |
|
public static ForkJoinPool commonPool() { |
2564 |
< |
return commonPool; // cannot be null (if so, a static init error) |
2564 |
> |
// assert common != null : "static init error"; |
2565 |
> |
return common; |
2566 |
|
} |
2567 |
|
|
2568 |
|
// Execution methods |
2734 |
|
* @return the targeted parallelism level of this pool |
2735 |
|
*/ |
2736 |
|
public int getParallelism() { |
2737 |
< |
return parallelism; |
2737 |
> |
return config & SMASK; |
2738 |
|
} |
2739 |
|
|
2740 |
|
/** |
2743 |
|
* @return the targeted parallelism level of the common pool |
2744 |
|
*/ |
2745 |
|
public static int getCommonPoolParallelism() { |
2746 |
< |
return commonPoolParallelism; |
2746 |
> |
return commonParallelism; |
2747 |
|
} |
2748 |
|
|
2749 |
|
/** |
2755 |
|
* @return the number of worker threads |
2756 |
|
*/ |
2757 |
|
public int getPoolSize() { |
2758 |
< |
return parallelism + (short)(ctl >>> TC_SHIFT); |
2758 |
> |
return (config & SMASK) + (short)(ctl >>> TC_SHIFT); |
2759 |
|
} |
2760 |
|
|
2761 |
|
/** |
2765 |
|
* @return {@code true} if this pool uses async mode |
2766 |
|
*/ |
2767 |
|
public boolean getAsyncMode() { |
2768 |
< |
return localMode != 0; |
2768 |
> |
return (config >>> 16) == FIFO_QUEUE; |
2769 |
|
} |
2770 |
|
|
2771 |
|
/** |
2796 |
|
* @return the number of active threads |
2797 |
|
*/ |
2798 |
|
public int getActiveThreadCount() { |
2799 |
< |
int r = parallelism + (int)(ctl >> AC_SHIFT); |
2799 |
> |
int r = (config & SMASK) + (int)(ctl >> AC_SHIFT); |
2800 |
|
return (r <= 0) ? 0 : r; // suppress momentarily negative values |
2801 |
|
} |
2802 |
|
|
2812 |
|
* @return {@code true} if all threads are currently idle |
2813 |
|
*/ |
2814 |
|
public boolean isQuiescent() { |
2815 |
< |
return (int)(ctl >> AC_SHIFT) + parallelism == 0; |
2815 |
> |
return (int)(ctl >> AC_SHIFT) + (config & SMASK) == 0; |
2816 |
|
} |
2817 |
|
|
2818 |
|
/** |
2889 |
|
WorkQueue[] ws; WorkQueue w; |
2890 |
|
if ((ws = workQueues) != null) { |
2891 |
|
for (int i = 0; i < ws.length; i += 2) { |
2892 |
< |
if ((w = ws[i]) != null && w.queueSize() != 0) |
2892 |
> |
if ((w = ws[i]) != null && !w.isEmpty()) |
2893 |
|
return true; |
2894 |
|
} |
2895 |
|
} |
2975 |
|
} |
2976 |
|
} |
2977 |
|
} |
2978 |
< |
int pc = parallelism; |
2978 |
> |
int pc = (config & SMASK); |
2979 |
|
int tc = pc + (short)(c >>> TC_SHIFT); |
2980 |
|
int ac = pc + (int)(c >> AC_SHIFT); |
2981 |
|
if (ac < 0) // ignore transient negative |
3048 |
|
public boolean isTerminated() { |
3049 |
|
long c = ctl; |
3050 |
|
return ((c & STOP_BIT) != 0L && |
3051 |
< |
(short)(c >>> TC_SHIFT) == -parallelism); |
3051 |
> |
(short)(c >>> TC_SHIFT) == -(config & SMASK)); |
3052 |
|
} |
3053 |
|
|
3054 |
|
/** |
3056 |
|
* commenced but not yet completed. This method may be useful for |
3057 |
|
* debugging. A return of {@code true} reported a sufficient |
3058 |
|
* period after shutdown may indicate that submitted tasks have |
3059 |
< |
* ignored or suppressed interruption, or are waiting for IO, |
3059 |
> |
* ignored or suppressed interruption, or are waiting for I/O, |
3060 |
|
* causing this executor not to properly terminate. (See the |
3061 |
|
* advisory notes for class {@link ForkJoinTask} stating that |
3062 |
|
* tasks should not normally entail blocking operations. But if |
3067 |
|
public boolean isTerminating() { |
3068 |
|
long c = ctl; |
3069 |
|
return ((c & STOP_BIT) != 0L && |
3070 |
< |
(short)(c >>> TC_SHIFT) != -parallelism); |
3070 |
> |
(short)(c >>> TC_SHIFT) != -(config & SMASK)); |
3071 |
|
} |
3072 |
|
|
3073 |
|
/** |
3082 |
|
/** |
3083 |
|
* Blocks until all tasks have completed execution after a |
3084 |
|
* shutdown request, or the timeout occurs, or the current thread |
3085 |
< |
* is interrupted, whichever happens first. Note that the {@link |
3086 |
< |
* #commonPool()} never terminates until program shutdown so |
3087 |
< |
* this method will always time out. |
3085 |
> |
* is interrupted, whichever happens first. Because the {@link |
3086 |
> |
* #commonPool()} never terminates until program shutdown, when |
3087 |
> |
* applied to the common pool, this method is equivalent to {@link |
3088 |
> |
* #awaitQuiescence} but always returns {@code false}. |
3089 |
|
* |
3090 |
|
* @param timeout the maximum time to wait |
3091 |
|
* @param unit the time unit of the timeout argument |
3095 |
|
*/ |
3096 |
|
public boolean awaitTermination(long timeout, TimeUnit unit) |
3097 |
|
throws InterruptedException { |
3098 |
+ |
if (Thread.interrupted()) |
3099 |
+ |
throw new InterruptedException(); |
3100 |
+ |
if (this == common) { |
3101 |
+ |
awaitQuiescence(timeout, unit); |
3102 |
+ |
return false; |
3103 |
+ |
} |
3104 |
|
long nanos = unit.toNanos(timeout); |
3105 |
|
if (isTerminated()) |
3106 |
|
return true; |
3120 |
|
} |
3121 |
|
|
3122 |
|
/** |
3123 |
+ |
* If called by a ForkJoinTask operating in this pool, equivalent |
3124 |
+ |
* in effect to {@link ForkJoinTask#helpQuiesce}. Otherwise, |
3125 |
+ |
* waits and/or attempts to assist performing tasks until this |
3126 |
+ |
* pool {@link #isQuiescent} or the indicated timeout elapses. |
3127 |
+ |
* |
3128 |
+ |
* @param timeout the maximum time to wait |
3129 |
+ |
* @param unit the time unit of the timeout argument |
3130 |
+ |
* @return {@code true} if quiescent; {@code false} if the |
3131 |
+ |
* timeout elapsed. |
3132 |
+ |
*/ |
3133 |
+ |
public boolean awaitQuiescence(long timeout, TimeUnit unit) { |
3134 |
+ |
long nanos = unit.toNanos(timeout); |
3135 |
+ |
ForkJoinWorkerThread wt; |
3136 |
+ |
Thread thread = Thread.currentThread(); |
3137 |
+ |
if ((thread instanceof ForkJoinWorkerThread) && |
3138 |
+ |
(wt = (ForkJoinWorkerThread)thread).pool == this) { |
3139 |
+ |
helpQuiescePool(wt.workQueue); |
3140 |
+ |
return true; |
3141 |
+ |
} |
3142 |
+ |
long startTime = System.nanoTime(); |
3143 |
+ |
WorkQueue[] ws; |
3144 |
+ |
int r = 0, m; |
3145 |
+ |
boolean found = true; |
3146 |
+ |
while (!isQuiescent() && (ws = workQueues) != null && |
3147 |
+ |
(m = ws.length - 1) >= 0) { |
3148 |
+ |
if (!found) { |
3149 |
+ |
if ((System.nanoTime() - startTime) > nanos) |
3150 |
+ |
return false; |
3151 |
+ |
Thread.yield(); // cannot block |
3152 |
+ |
} |
3153 |
+ |
found = false; |
3154 |
+ |
for (int j = (m + 1) << 2; j >= 0; --j) { |
3155 |
+ |
ForkJoinTask<?> t; WorkQueue q; int b; |
3156 |
+ |
if ((q = ws[r++ & m]) != null && (b = q.base) - q.top < 0) { |
3157 |
+ |
found = true; |
3158 |
+ |
if ((t = q.pollAt(b)) != null) { |
3159 |
+ |
if (q.base - q.top < 0) |
3160 |
+ |
signalWork(q); |
3161 |
+ |
t.doExec(); |
3162 |
+ |
} |
3163 |
+ |
break; |
3164 |
+ |
} |
3165 |
+ |
} |
3166 |
+ |
} |
3167 |
+ |
return true; |
3168 |
+ |
} |
3169 |
+ |
|
3170 |
+ |
/** |
3171 |
+ |
* Waits and/or attempts to assist performing tasks indefinitely |
3172 |
+ |
* until the {@link #commonPool()} {@link #isQuiescent} |
3173 |
+ |
*/ |
3174 |
+ |
public static void quiesceCommonPool() { |
3175 |
+ |
common.awaitQuiescence(Long.MAX_VALUE, TimeUnit.NANOSECONDS); |
3176 |
+ |
} |
3177 |
+ |
|
3178 |
+ |
/** |
3179 |
|
* Interface for extending managed parallelism for tasks running |
3180 |
|
* in {@link ForkJoinPool}s. |
3181 |
|
* |
3274 |
|
if (t instanceof ForkJoinWorkerThread) { |
3275 |
|
ForkJoinPool p = ((ForkJoinWorkerThread)t).pool; |
3276 |
|
while (!blocker.isReleasable()) { // variant of helpSignal |
3277 |
< |
WorkQueue[] ws; WorkQueue q; int m, n; |
3277 |
> |
WorkQueue[] ws; WorkQueue q; int m, u; |
3278 |
|
if ((ws = p.workQueues) != null && (m = ws.length - 1) >= 0) { |
3279 |
|
for (int i = 0; i <= m; ++i) { |
3280 |
|
if (blocker.isReleasable()) |
3281 |
|
return; |
3282 |
< |
if ((q = ws[i]) != null && (n = q.queueSize()) > 0) { |
3283 |
< |
p.signalWork(q, n); |
3284 |
< |
if ((int)(p.ctl >> AC_SHIFT) >= 0) |
3282 |
> |
if ((q = ws[i]) != null && q.base - q.top < 0) { |
3283 |
> |
p.signalWork(q); |
3284 |
> |
if ((u = (int)(p.ctl >>> 32)) >= 0 || |
3285 |
> |
(u >> UAC_SHIFT) >= 0) |
3286 |
|
break; |
3287 |
|
} |
3288 |
|
} |
3328 |
|
private static final long QLOCK; |
3329 |
|
|
3330 |
|
static { |
3182 |
– |
// Establish common pool parameters |
3183 |
– |
// TBD: limit or report ignored exceptions? |
3184 |
– |
|
3185 |
– |
int par = 0; |
3186 |
– |
ForkJoinWorkerThreadFactory fac = null; |
3187 |
– |
Thread.UncaughtExceptionHandler handler = null; |
3188 |
– |
try { |
3189 |
– |
String pp = System.getProperty(propPrefix + "parallelism"); |
3190 |
– |
String hp = System.getProperty(propPrefix + "exceptionHandler"); |
3191 |
– |
String fp = System.getProperty(propPrefix + "threadFactory"); |
3192 |
– |
if (fp != null) |
3193 |
– |
fac = ((ForkJoinWorkerThreadFactory)ClassLoader. |
3194 |
– |
getSystemClassLoader().loadClass(fp).newInstance()); |
3195 |
– |
if (hp != null) |
3196 |
– |
handler = ((Thread.UncaughtExceptionHandler)ClassLoader. |
3197 |
– |
getSystemClassLoader().loadClass(hp).newInstance()); |
3198 |
– |
if (pp != null) |
3199 |
– |
par = Integer.parseInt(pp); |
3200 |
– |
} catch (Exception ignore) { |
3201 |
– |
} |
3202 |
– |
|
3331 |
|
int s; // initialize field offsets for CAS etc |
3332 |
|
try { |
3333 |
|
U = getUnsafe(); |
3356 |
|
if ((s & (s-1)) != 0) |
3357 |
|
throw new Error("data type scale not a power of two"); |
3358 |
|
|
3359 |
+ |
submitters = new ThreadLocal<Submitter>(); |
3360 |
+ |
ForkJoinWorkerThreadFactory fac = defaultForkJoinWorkerThreadFactory = |
3361 |
+ |
new DefaultForkJoinWorkerThreadFactory(); |
3362 |
+ |
modifyThreadPermission = new RuntimePermission("modifyThread"); |
3363 |
+ |
|
3364 |
|
/* |
3365 |
< |
* For extra caution, computations to set up pool state are |
3366 |
< |
* here; the constructor just assigns these values to fields. |
3365 |
> |
* Establish common pool parameters. For extra caution, |
3366 |
> |
* computations to set up common pool state are here; the |
3367 |
> |
* constructor just assigns these values to fields. |
3368 |
|
*/ |
3369 |
< |
ForkJoinWorkerThreadFactory defaultFac = |
3370 |
< |
defaultForkJoinWorkerThreadFactory = |
3371 |
< |
new DefaultForkJoinWorkerThreadFactory(); |
3372 |
< |
if (fac == null) |
3373 |
< |
fac = defaultFac; |
3369 |
> |
|
3370 |
> |
int par = 0; |
3371 |
> |
Thread.UncaughtExceptionHandler handler = null; |
3372 |
> |
try { // TBD: limit or report ignored exceptions? |
3373 |
> |
String pp = System.getProperty |
3374 |
> |
("java.util.concurrent.ForkJoinPool.common.parallelism"); |
3375 |
> |
String hp = System.getProperty |
3376 |
> |
("java.util.concurrent.ForkJoinPool.common.exceptionHandler"); |
3377 |
> |
String fp = System.getProperty |
3378 |
> |
("java.util.concurrent.ForkJoinPool.common.threadFactory"); |
3379 |
> |
if (fp != null) |
3380 |
> |
fac = ((ForkJoinWorkerThreadFactory)ClassLoader. |
3381 |
> |
getSystemClassLoader().loadClass(fp).newInstance()); |
3382 |
> |
if (hp != null) |
3383 |
> |
handler = ((Thread.UncaughtExceptionHandler)ClassLoader. |
3384 |
> |
getSystemClassLoader().loadClass(hp).newInstance()); |
3385 |
> |
if (pp != null) |
3386 |
> |
par = Integer.parseInt(pp); |
3387 |
> |
} catch (Exception ignore) { |
3388 |
> |
} |
3389 |
> |
|
3390 |
|
if (par <= 0) |
3391 |
|
par = Runtime.getRuntime().availableProcessors(); |
3392 |
|
if (par > MAX_CAP) |
3393 |
|
par = MAX_CAP; |
3394 |
+ |
commonParallelism = par; |
3395 |
|
long np = (long)(-par); // precompute initial ctl value |
3396 |
|
long ct = ((np << AC_SHIFT) & AC_MASK) | ((np << TC_SHIFT) & TC_MASK); |
3397 |
|
|
3398 |
< |
commonPoolParallelism = par; |
3248 |
< |
commonPool = new ForkJoinPool(par, ct, fac, handler); |
3249 |
< |
modifyThreadPermission = new RuntimePermission("modifyThread"); |
3250 |
< |
submitters = new ThreadLocal<Submitter>(); |
3398 |
> |
common = new ForkJoinPool(par, ct, fac, handler); |
3399 |
|
} |
3400 |
|
|
3401 |
|
/** |