--- jsr166/src/jsr166y/ForkJoinPool.java 2010/11/29 20:58:06 1.90 +++ jsr166/src/jsr166y/ForkJoinPool.java 2011/02/22 00:39:31 1.91 @@ -11,6 +11,7 @@ import java.util.Arrays; import java.util.Collection; import java.util.Collections; import java.util.List; +import java.util.Random; import java.util.concurrent.AbstractExecutorService; import java.util.concurrent.Callable; import java.util.concurrent.ExecutorService; @@ -22,6 +23,7 @@ import java.util.concurrent.TimeoutExcep import java.util.concurrent.atomic.AtomicInteger; import java.util.concurrent.locks.LockSupport; import java.util.concurrent.locks.ReentrantLock; +import java.util.concurrent.locks.Condition; /** * An {@link ExecutorService} for running {@link ForkJoinTask}s. @@ -129,239 +131,203 @@ public class ForkJoinPool extends Abstra * set of worker threads: Submissions from non-FJ threads enter * into a submission queue. Workers take these tasks and typically * split them into subtasks that may be stolen by other workers. - * The main work-stealing mechanics implemented in class - * ForkJoinWorkerThread give first priority to processing tasks - * from their own queues (LIFO or FIFO, depending on mode), then - * to randomized FIFO steals of tasks in other worker queues, and - * lastly to new submissions. These mechanics do not consider - * affinities, loads, cache localities, etc, so rarely provide the - * best possible performance on a given machine, but portably - * provide good throughput by averaging over these factors. - * (Further, even if we did try to use such information, we do not - * usually have a basis for exploiting it. For example, some sets - * of tasks profit from cache affinities, but others are harmed by - * cache pollution effects.) - * - * Beyond work-stealing support and essential bookkeeping, the - * main responsibility of this framework is to take actions when - * one worker is waiting to join a task stolen (or always held by) - * another. Because we are multiplexing many tasks on to a pool - * of workers, we can't just let them block (as in Thread.join). - * We also cannot just reassign the joiner's run-time stack with - * another and replace it later, which would be a form of - * "continuation", that even if possible is not necessarily a good - * idea. Given that the creation costs of most threads on most - * systems mainly surrounds setting up runtime stacks, thread - * creation and switching is usually not much more expensive than - * stack creation and switching, and is more flexible). Instead we + * Preference rules give first priority to processing tasks from + * their own queues (LIFO or FIFO, depending on mode), then to + * randomized FIFO steals of tasks in other worker queues, and + * lastly to new submissions. + * + * The main throughput advantages of work-stealing stem from + * decentralized control -- workers mostly take tasks from + * themselves or each other. We cannot negate this in the + * implementation of other management responsibilities. The main + * tactic for avoiding bottlenecks is packing nearly all + * essentially atomic control state into a single 64bit volatile + * variable ("ctl"). This variable is read on the order of 10-100 + * times as often as it is modified (always via CAS). (There is + * some additional control state, for example variable "shutdown" + * for which we can cope with uncoordinated updates.) This + * streamlines synchronization and control at the expense of messy + * constructions needed to repack status bits upon updates. + * Updates tend not to contend with each other except during + * bursts while submitted tasks begin or end. In some cases when + * they do contend, threads can instead do something else + * (usually, scan for tesks) until contention subsides. + * + * To enable packing, we restrict maximum parallelism to (1<<15)-1 + * (which is far in excess of normal operating range) to allow + * ids, counts, and their negations (used for thresholding) to fit + * into 16bit fields. + * + * Recording Workers. Workers are recorded in the "workers" array + * that is created upon pool construction and expanded if (rarely) + * necessary. This is an array as opposed to some other data + * structure to support index-based random steals by workers. + * Updates to the array recording new workers and unrecording + * terminated ones are protected from each other by a seqLock + * (scanGuard) but the array is otherwise concurrently readable, + * and accessed directly by workers. To simplify index-based + * operations, the array size is always a power of two, and all + * readers must tolerate null slots. To avoid flailing during + * start-up, the array is presized to hold twice #parallelism + * workers (which is unlikely to need further resizing during + * execution). But to avoid dealing with so many null slots, + * variable scanGuard includes a mask for the nearest power of two + * that contains all current workers. All worker thread creation + * is on-demand, triggered by task submissions, replacement of + * terminated workers, and/or compensation for blocked + * workers. However, all other support code is set up to work with + * other policies. To ensure that we do not hold on to worker + * references that would prevent GC, ALL accesses to workers are + * via indices into the workers array (which is one source of some + * of the messy code constructions here). In essence, the workers + * array serves as a weak reference mechanism. Thus for example + * the wait queue field of ctl stores worker indices, not worker + * references. Access to the workers in associated methods (for + * example signalWork) must both index-check and null-check the + * IDs. All such accesses ignore bad IDs by returning out early + * from what they are doing, since this can only be associated + * with termination, in which case it is OK to give up. + * + * All uses of the workers array, as well as queue arrays, check + * that the array is non-null (even if previously non-null). This + * allows nulling during termination, which is currently not + * necessary, but remains an option for resource-revocation-based + * shutdown schemes. + * + * Wait Queuing. Unlike HPC work-stealing frameworks, we cannot + * let workers spin indefinitely scanning for tasks when none are + * can be immediately found, and we cannot start/resume workers + * unless there appear to be tasks available. On the other hand, + * we must quickly prod them into action when new tasks are + * submitted or generated. We park/unpark workers after placing + * in an event wait queue when they cannot find work. This "queue" + * is actually a simple Treiber stack, headed by the "id" field of + * ctl, plus a 15bit counter value to both wake up waiters (by + * advancing their count) and avoid ABA effects. Successors are + * held in worker field "nextWait". Queuing deals with several + * intrinsic races, mainly that a task-producing thread can miss + * seeing (and signalling) another thread that gave up looking for + * work but has not yet entered the wait queue. We solve this by + * requiring a full sweep of all workers both before (in scan()) + * and after (in awaitWork()) a newly waiting worker is added to + * the wait queue. During a rescan, the worker might release some + * other queued worker rather than itself, which has the same net + * effect. + * + * Signalling. We create or wake up workers only when there + * appears to be at least one task they might be able to find and + * execute. When a submission is added or another worker adds a + * task to a queue that previously had two or fewer tasks, they + * signal waiting workers (or trigger creation of new ones if + * fewer than the given parallelism level -- see signalWork). + * These primary signals are buttressed by signals during rescans + * as well as those performed when a worker steals a task and + * notices that there are more tasks too; together these cover the + * signals needed in cases when more than two tasks are pushed + * but untaken. + * + * Trimming workers. To release resources after periods of lack of + * use, a worker starting to wait when the pool is quiescent will + * time out and terminate if the pool has remained quiescent for + * SHRINK_RATE nanosecs. + * + * Submissions. External submissions are maintained in an + * array-based queue that is structured identically to + * ForkJoinWorkerThread queues (which see) except for the use of + * submissionLock in method addSubmission. Unlike worker queues, + * multiple external threads can add new submissions. + * + * Compensation. Beyond work-stealing support and lifecycle + * control, the main responsibility of this framework is to take + * actions when one worker is waiting to join a task stolen (or + * always held by) another. Because we are multiplexing many + * tasks on to a pool of workers, we can't just let them block (as + * in Thread.join). We also cannot just reassign the joiner's + * run-time stack with another and replace it later, which would + * be a form of "continuation", that even if possible is not + * necessarily a good idea since we sometimes need both an + * unblocked task and its continuation to progress. Instead we * combine two tactics: * * Helping: Arranging for the joiner to execute some task that it * would be running if the steal had not occurred. Method - * ForkJoinWorkerThread.helpJoinTask tracks joining->stealing + * ForkJoinWorkerThread.joinTask tracks joining->stealing * links to try to find such a task. * * Compensating: Unless there are already enough live threads, - * method helpMaintainParallelism() may create or - * re-activate a spare thread to compensate for blocked - * joiners until they unblock. - * - * It is impossible to keep exactly the target (parallelism) - * number of threads running at any given time. Determining - * existence of conservatively safe helping targets, the - * availability of already-created spares, and the apparent need - * to create new spares are all racy and require heuristic - * guidance, so we rely on multiple retries of each. Compensation - * occurs in slow-motion. It is triggered only upon timeouts of - * Object.wait used for joins. This reduces poor decisions that - * would otherwise be made when threads are waiting for others - * that are stalled because of unrelated activities such as - * garbage collection. + * method tryPreBlock() may create or re-activate a spare + * thread to compensate for blocked joiners until they + * unblock. * * The ManagedBlocker extension API can't use helping so relies * only on compensation in method awaitBlocker. * - * The main throughput advantages of work-stealing stem from - * decentralized control -- workers mostly steal tasks from each - * other. We do not want to negate this by creating bottlenecks - * implementing other management responsibilities. So we use a - * collection of techniques that avoid, reduce, or cope well with - * contention. These entail several instances of bit-packing into - * CASable fields to maintain only the minimally required - * atomicity. To enable such packing, we restrict maximum - * parallelism to (1<<15)-1 (enabling twice this (to accommodate - * unbalanced increments and decrements) to fit into a 16 bit - * field, which is far in excess of normal operating range. Even - * though updates to some of these bookkeeping fields do sometimes - * contend with each other, they don't normally cache-contend with - * updates to others enough to warrant memory padding or - * isolation. So they are all held as fields of ForkJoinPool - * objects. The main capabilities are as follows: - * - * 1. Creating and removing workers. Workers are recorded in the - * "workers" array. This is an array as opposed to some other data - * structure to support index-based random steals by workers. - * Updates to the array recording new workers and unrecording - * terminated ones are protected from each other by a lock - * (workerLock) but the array is otherwise concurrently readable, - * and accessed directly by workers. To simplify index-based - * operations, the array size is always a power of two, and all - * readers must tolerate null slots. Currently, all worker thread - * creation is on-demand, triggered by task submissions, - * replacement of terminated workers, and/or compensation for - * blocked workers. However, all other support code is set up to - * work with other policies. - * - * To ensure that we do not hold on to worker references that - * would prevent GC, ALL accesses to workers are via indices into - * the workers array (which is one source of some of the unusual - * code constructions here). In essence, the workers array serves - * as a WeakReference mechanism. Thus for example the event queue - * stores worker indices, not worker references. Access to the - * workers in associated methods (for example releaseEventWaiters) - * must both index-check and null-check the IDs. All such accesses - * ignore bad IDs by returning out early from what they are doing, - * since this can only be associated with shutdown, in which case - * it is OK to give up. On termination, we just clobber these - * data structures without trying to use them. - * - * 2. Bookkeeping for dynamically adding and removing workers. We - * aim to approximately maintain the given level of parallelism. - * When some workers are known to be blocked (on joins or via - * ManagedBlocker), we may create or resume others to take their - * place until they unblock (see below). Implementing this - * requires counts of the number of "running" threads (i.e., those - * that are neither blocked nor artificially suspended) as well as - * the total number. These two values are packed into one field, - * "workerCounts" because we need accurate snapshots when deciding - * to create, resume or suspend. Note however that the - * correspondence of these counts to reality is not guaranteed. In - * particular updates for unblocked threads may lag until they - * actually wake up. - * - * 3. Maintaining global run state. The run state of the pool - * consists of a runLevel (SHUTDOWN, TERMINATING, etc) similar to - * those in other Executor implementations, as well as a count of - * "active" workers -- those that are, or soon will be, or - * recently were executing tasks. The runLevel and active count - * are packed together in order to correctly trigger shutdown and - * termination. Without care, active counts can be subject to very - * high contention. We substantially reduce this contention by - * relaxing update rules. A worker must claim active status - * prospectively, by activating if it sees that a submitted or - * stealable task exists (it may find after activating that the - * task no longer exists). It stays active while processing this - * task (if it exists) and any other local subtasks it produces, - * until it cannot find any other tasks. It then tries - * inactivating (see method preStep), but upon update contention - * instead scans for more tasks, later retrying inactivation if it - * doesn't find any. - * - * 4. Managing idle workers waiting for tasks. We cannot let - * workers spin indefinitely scanning for tasks when none are - * available. On the other hand, we must quickly prod them into - * action when new tasks are submitted or generated. We - * park/unpark these idle workers using an event-count scheme. - * Field eventCount is incremented upon events that may enable - * workers that previously could not find a task to now find one: - * Submission of a new task to the pool, or another worker pushing - * a task onto a previously empty queue. (We also use this - * mechanism for configuration and termination actions that - * require wakeups of idle workers). Each worker maintains its - * last known event count, and blocks when a scan for work did not - * find a task AND its lastEventCount matches the current - * eventCount. Waiting idle workers are recorded in a variant of - * Treiber stack headed by field eventWaiters which, when nonzero, - * encodes the thread index and count awaited for by the worker - * thread most recently calling eventSync. This thread in turn has - * a record (field nextEventWaiter) for the next waiting worker. - * In addition to allowing simpler decisions about need for - * wakeup, the event count bits in eventWaiters serve the role of - * tags to avoid ABA errors in Treiber stacks. Upon any wakeup, - * released threads also try to release at most two others. The - * net effect is a tree-like diffusion of signals, where released - * threads (and possibly others) help with unparks. To further - * reduce contention effects a bit, failed CASes to increment - * field eventCount are tolerated without retries in signalWork. - * Conceptually they are merged into the same event, which is OK - * when their only purpose is to enable workers to scan for work. - * - * 5. Managing suspension of extra workers. When a worker notices - * (usually upon timeout of a wait()) that there are too few - * running threads, we may create a new thread to maintain - * parallelism level, or at least avoid starvation. Usually, extra - * threads are needed for only very short periods, yet join - * dependencies are such that we sometimes need them in - * bursts. Rather than create new threads each time this happens, - * we suspend no-longer-needed extra ones as "spares". For most - * purposes, we don't distinguish "extra" spare threads from - * normal "core" threads: On each call to preStep (the only point - * at which we can do this) a worker checks to see if there are - * now too many running workers, and if so, suspends itself. - * Method helpMaintainParallelism looks for suspended threads to - * resume before considering creating a new replacement. The - * spares themselves are encoded on another variant of a Treiber - * Stack, headed at field "spareWaiters". Note that the use of - * spares is intrinsically racy. One thread may become a spare at - * about the same time as another is needlessly being created. We - * counteract this and related slop in part by requiring resumed - * spares to immediately recheck (in preStep) to see whether they - * should re-suspend. - * - * 6. Killing off unneeded workers. A timeout mechanism is used to - * shed unused workers: The oldest (first) event queue waiter uses - * a timed rather than hard wait. When this wait times out without - * a normal wakeup, it tries to shutdown any one (for convenience - * the newest) other spare or event waiter via - * tryShutdownUnusedWorker. This eventually reduces the number of - * worker threads to a minimum of one after a long enough period - * without use. - * - * 7. Deciding when to create new workers. The main dynamic - * control in this class is deciding when to create extra threads - * in method helpMaintainParallelism. We would like to keep - * exactly #parallelism threads running, which is an impossible - * task. We always need to create one when the number of running - * threads would become zero and all workers are busy. Beyond - * this, we must rely on heuristics that work well in the - * presence of transient phenomena such as GC stalls, dynamic - * compilation, and wake-up lags. These transients are extremely - * common -- we are normally trying to fully saturate the CPUs on - * a machine, so almost any activity other than running tasks - * impedes accuracy. Our main defense is to allow parallelism to - * lapse for a while during joins, and use a timeout to see if, - * after the resulting settling, there is still a need for - * additional workers. This also better copes with the fact that - * some of the methods in this class tend to never become compiled - * (but are interpreted), so some components of the entire set of - * controls might execute 100 times faster than others. And - * similarly for cases where the apparent lack of work is just due - * to GC stalls and other transient system activity. + * It is impossible to keep exactly the target parallelism number + * of threads running at any given time. Determining the + * existence of conservatively safe helping targets, the + * availability of already-created spares, and the apparent need + * to create new spares are all racy and require heuristic + * guidance, so we rely on multiple retries of each. Currently, + * in keeping with on-demand signalling policy, we compensate only + * if blocking would leave less than one active (non-waiting, + * non-blocked) worker. Additionally, to avoid some false alarms + * due to GC, lagging counters, system activity, etc, compensated + * blocking for joins is only attempted after a number of rechecks + * proportional to the current apparent deficit (where retries are + * interspersed with Thread.yield, for good citizenship). The + * variable blockedCount, incremented before blocking and + * decremented after, is sometimes needed to distinguish cases of + * waiting for work vs blocking on joins or other managed sync, + * but both the cases are equivalent for most pool control, so we + * can update non-atomically. (Additionally, contention on + * blockedCount alleviates some contention on ctl). + * + * Shutdown and Termination. A call to shutdownNow atomically sets + * the ctl stop bit and then (non-atomically) sets each workers + * "terminate" status, cancels all unprocessed tasks, and wakes up + * all waiting workers. Detecting whether termination should + * commence after a non-abrupt shutdown() call requires more work + * and bookkeeping. We need consensus about quiesence (i.e., that + * there is no more work) which is reflected in active counts so + * long as there are no current blockers, as well as possible + * re-evaluations during independent changes in blocking or + * quiescing workers. * - * Beware that there is a lot of representation-level coupling + * Style notes: There is a lot of representation-level coupling * among classes ForkJoinPool, ForkJoinWorkerThread, and - * ForkJoinTask. For example, direct access to "workers" array by + * ForkJoinTask. Most fields of ForkJoinWorkerThread maintain + * data structures managed by ForkJoinPool, so are directly + * accessed. Conversely we allow access to "workers" array by * workers, and direct access to ForkJoinTask.status by both * ForkJoinPool and ForkJoinWorkerThread. There is little point * trying to reduce this, since any associated future changes in * representations will need to be accompanied by algorithmic - * changes anyway. - * - * Style notes: There are lots of inline assignments (of form - * "while ((local = field) != 0)") which are usually the simplest - * way to ensure the required read orderings (which are sometimes - * critical). Also several occurrences of the unusual "do {} - * while (!cas...)" which is the simplest way to force an update of - * a CAS'ed variable. There are also other coding oddities that - * help some methods perform reasonably even when interpreted (not - * compiled), at the expense of some messy constructions that - * reduce byte code counts. - * - * The order of declarations in this file is: (1) statics (2) - * fields (along with constants used when unpacking some of them) - * (3) internal control methods (4) callbacks and other support - * for ForkJoinTask and ForkJoinWorkerThread classes, (5) exported - * methods (plus a few little helpers). + * changes anyway. All together, these low-level implementation + * choices produce as much as a factor of 4 performance + * improvement compared to naive implementations, and enable the + * processing of billions of tasks per second, at the expense of + * some ugliness. + * + * Methods signalWork() and scan() are the main bottlenecks so are + * especially heavily micro-optimized/mangled. There are lots of + * inline assignments (of form "while ((local = field) != 0)") + * which are usually the simplest way to ensure the required read + * orderings (which are sometimes critical). This leads to a + * "C"-like style of listing declarations of these locals at the + * heads of methods or blocks. There are several occurrences of + * the unusual "do {} while (!cas...)" which is the simplest way + * to force an update of a CAS'ed variable. There are also other + * coding oddities that help some methods perform reasonably even + * when interpreted (not compiled). + * + * The order of declarations in this file is: (1) declarations of + * statics (2) fields (along with constants used when unpacking + * some of them), listed in an order that tends to reduce + * contention among them a bit under most JVMs. (3) internal + * control methods (4) callbacks and other support for + * ForkJoinTask and ForkJoinWorkerThread classes, (5) exported + * methods (plus a few little helpers). (6) static block + * initializing all statics in a minimally dependent order. */ /** @@ -396,15 +362,13 @@ public class ForkJoinPool extends Abstra * overridden in ForkJoinPool constructors. */ public static final ForkJoinWorkerThreadFactory - defaultForkJoinWorkerThreadFactory = - new DefaultForkJoinWorkerThreadFactory(); + defaultForkJoinWorkerThreadFactory; /** * Permission required for callers of methods that may start or * kill threads. */ - private static final RuntimePermission modifyThreadPermission = - new RuntimePermission("modifyThread"); + private static final RuntimePermission modifyThreadPermission; /** * If there is a security manager, makes sure caller has @@ -419,63 +383,59 @@ public class ForkJoinPool extends Abstra /** * Generator for assigning sequence numbers as pool names. */ - private static final AtomicInteger poolNumberGenerator = - new AtomicInteger(); + private static final AtomicInteger poolNumberGenerator; /** - * The time to block in a join (see awaitJoin) before checking if - * a new worker should be (re)started to maintain parallelism - * level. The value should be short enough to maintain global - * responsiveness and progress but long enough to avoid - * counterproductive firings during GC stalls or unrelated system - * activity, and to not bog down systems with continual re-firings - * on GCs or legitimately long waits. + * Generator for initial random seeds for worker victim + * selection. This is used only to create initial seeds. Random + * steals use a cheaper xorshift generator per steal attempt. We + * don't expect much contention on seedGenerator, so just use a + * plain Random. */ - private static final long JOIN_TIMEOUT_MILLIS = 250L; // 4 per second + static final Random workerSeedGenerator; /** - * The wakeup interval (in nanoseconds) for the oldest worker - * waiting for an event to invoke tryShutdownUnusedWorker to - * shrink the number of workers. The exact value does not matter - * too much. It must be short enough to release resources during - * sustained periods of idleness, but not so short that threads - * are continually re-created. + * Array holding all worker threads in the pool. Initialized upon + * construction. Array size must be a power of two. Updates and + * replacements are protected by scanGuard, but the array is + * always kept in a consistent enough state to be randomly + * accessed without locking by workers performing work-stealing, + * as well as other traversal-based methods in this class, so long + * as reads memory-acquire by first reading ctl. All readers must + * tolerate that some array slots may be null. */ - private static final long SHRINK_RATE_NANOS = - 30L * 1000L * 1000L * 1000L; // 2 per minute + ForkJoinWorkerThread[] workers; /** - * Absolute bound for parallelism level. Twice this number plus - * one (i.e., 0xfff) must fit into a 16bit field to enable - * word-packing for some counts and indices. + * Initial size for submission queue array. Must be a power of + * two. In many applications, these always stay small so we use a + * small initial cap. */ - private static final int MAX_WORKERS = 0x7fff; + private static final int INITIAL_QUEUE_CAPACITY = 8; /** - * Array holding all worker threads in the pool. Array size must - * be a power of two. Updates and replacements are protected by - * workerLock, but the array is always kept in a consistent enough - * state to be randomly accessed without locking by workers - * performing work-stealing, as well as other traversal-based - * methods in this class. All readers must tolerate that some - * array slots may be null. + * Maximum size for submission queue array. Must be a power of two + * less than or equal to 1 << (31 - width of array entry) to + * ensure lack of index wraparound, but is capped at a lower + * value to help users trap runaway computations. */ - volatile ForkJoinWorkerThread[] workers; + private static final int MAXIMUM_QUEUE_CAPACITY = 1 << 24; // 16M /** - * Queue for external submissions. + * Array serving as submission queue. Initialized upon construction. */ - private final LinkedTransferQueue> submissionQueue; + private ForkJoinTask[] submissionQueue; /** - * Lock protecting updates to workers array. + * Lock protecting submissions array for addSubmission */ - private final ReentrantLock workerLock; + private final ReentrantLock submissionLock; /** - * Latch released upon termination. + * Condition for awaitTermination, using submissionLock for + * convenience. */ - private final Phaser termination; + private final Condition termination; /** * Creation factory for worker threads. @@ -483,87 +443,101 @@ public class ForkJoinPool extends Abstra private final ForkJoinWorkerThreadFactory factory; /** - * Sum of per-thread steal counts, updated only when threads are - * idle or terminating. + * The uncaught exception handler used when any worker abruptly + * terminates. */ - private volatile long stealCount; + final Thread.UncaughtExceptionHandler ueh; /** - * Encoded record of top of Treiber stack of threads waiting for - * events. The top 32 bits contain the count being waited for. The - * bottom 16 bits contains one plus the pool index of waiting - * worker thread. (Bits 16-31 are unused.) + * Prefix for assigning names to worker threads */ - private volatile long eventWaiters; - - private static final int EVENT_COUNT_SHIFT = 32; - private static final int WAITER_ID_MASK = (1 << 16) - 1; + private final String workerNamePrefix; /** - * A counter for events that may wake up worker threads: - * - Submission of a new task to the pool - * - A worker pushing a task on an empty queue - * - termination + * Sum of per-thread steal counts, updated only when threads are + * idle or terminating. */ - private volatile int eventCount; + private volatile long stealCount; /** - * Encoded record of top of Treiber stack of spare threads waiting - * for resumption. The top 16 bits contain an arbitrary count to - * avoid ABA effects. The bottom 16bits contains one plus the pool - * index of waiting worker thread. - */ - private volatile int spareWaiters; - - private static final int SPARE_COUNT_SHIFT = 16; - private static final int SPARE_ID_MASK = (1 << 16) - 1; + * Main pool control -- a long packed with: + * AC: Number of active running workers minus target parallelism (16 bits) + * TC: Number of total workers minus target parallelism (16bits) + * ST: true if pool is terminating (1 bit) + * EC: the wait count of top waiting thread (15 bits) + * ID: ~poolIndex of top of Treiber stack of waiting threads (16 bits) + * + * When convenient, we can extract the upper 32 bits of counts and + * the lower 32 bits of queue state, u = (int)(ctl >>> 32) and e = + * (int)ctl. The ec field is never accessed alone, but always + * together with id and st. The offsets of counts by the target + * parallelism and the positionings of fields makes it possible to + * perform the most common checks via sign tests of fields: When + * ac is negative, there are not enough active workers, when tc is + * negative, there are not enough total workers, when id is + * negative, there is at least one waiting worker, and when e is + * negative, the pool is terminating. To deal with these possibly + * negative fields, we use casts in and out of "short" and/or + * signed shifts to maintain signedness. Note: AC_SHIFT is + * redundantly declared in ForkJoinWorkerThread in order to + * integrate a surplus-threads check. + */ + volatile long ctl; + + // bit positions/shifts for fields + private static final int AC_SHIFT = 48; + private static final int TC_SHIFT = 32; + private static final int ST_SHIFT = 31; + private static final int EC_SHIFT = 16; + + // bounds + private static final int MAX_ID = 0x7fff; // max poolIndex + private static final int SMASK = 0xffff; // mask short bits + private static final int SHORT_SIGN = 1 << 15; + private static final int INT_SIGN = 1 << 31; + + // masks + private static final long STOP_BIT = 0x0001L << ST_SHIFT; + private static final long AC_MASK = ((long)SMASK) << AC_SHIFT; + private static final long TC_MASK = ((long)SMASK) << TC_SHIFT; + + // units for incrementing and decrementing + private static final long TC_UNIT = 1L << TC_SHIFT; + private static final long AC_UNIT = 1L << AC_SHIFT; + + // masks and units for dealing with u = (int)(ctl >>> 32) + private static final int UAC_SHIFT = AC_SHIFT - 32; + private static final int UTC_SHIFT = TC_SHIFT - 32; + private static final int UAC_MASK = SMASK << UAC_SHIFT; + private static final int UTC_MASK = SMASK << UTC_SHIFT; + private static final int UAC_UNIT = 1 << UAC_SHIFT; + private static final int UTC_UNIT = 1 << UTC_SHIFT; + + // masks and units for dealing with e = (int)ctl + private static final int E_MASK = 0x7fffffff; // no STOP_BIT + private static final int EC_UNIT = 1 << EC_SHIFT; /** - * Lifecycle control. The low word contains the number of workers - * that are (probably) executing tasks. This value is atomically - * incremented before a worker gets a task to run, and decremented - * when a worker has no tasks and cannot find any. Bits 16-18 - * contain runLevel value. When all are zero, the pool is - * running. Level transitions are monotonic (running -> shutdown - * -> terminating -> terminated) so each transition adds a bit. - * These are bundled together to ensure consistent read for - * termination checks (i.e., that runLevel is at least SHUTDOWN - * and active threads is zero). - * - * Notes: Most direct CASes are dependent on these bitfield - * positions. Also, this field is non-private to enable direct - * performance-sensitive CASes in ForkJoinWorkerThread. + * The target parallelism level. */ - volatile int runState; - - // Note: The order among run level values matters. - private static final int RUNLEVEL_SHIFT = 16; - private static final int SHUTDOWN = 1 << RUNLEVEL_SHIFT; - private static final int TERMINATING = 1 << (RUNLEVEL_SHIFT + 1); - private static final int TERMINATED = 1 << (RUNLEVEL_SHIFT + 2); - private static final int ACTIVE_COUNT_MASK = (1 << RUNLEVEL_SHIFT) - 1; + final int parallelism; /** - * Holds number of total (i.e., created and not yet terminated) - * and running (i.e., not blocked on joins or other managed sync) - * threads, packed together to ensure consistent snapshot when - * making decisions about creating and suspending spare - * threads. Updated only by CAS. Note that adding a new worker - * requires incrementing both counts, since workers start off in - * running state. + * Index (mod submission queue length) of next element to take + * from submission queue. */ - private volatile int workerCounts; + volatile int queueBase; - private static final int TOTAL_COUNT_SHIFT = 16; - private static final int RUNNING_COUNT_MASK = (1 << TOTAL_COUNT_SHIFT) - 1; - private static final int ONE_RUNNING = 1; - private static final int ONE_TOTAL = 1 << TOTAL_COUNT_SHIFT; + /** + * Index (mod submission queue length) of next element to add + * in submission queue. + */ + int queueTop; /** - * The target parallelism level. - * Accessed directly by ForkJoinWorkerThreads. + * True when shutdown() has been called. */ - final int parallelism; + volatile boolean shutdown; /** * True if use local fifo, not default lifo, for local polling @@ -572,558 +546,652 @@ public class ForkJoinPool extends Abstra final boolean locallyFifo; /** - * The uncaught exception handler used when any worker abruptly - * terminates. + * The number of threads in ForkJoinWorkerThreads.helpQuiescePool. + * When non-zero, suppresses automatic shutdown when active + * counts become zero. */ - private final Thread.UncaughtExceptionHandler ueh; + volatile int quiescerCount; /** - * Pool number, just for assigning useful names to worker threads + * The number of threads blocked in join. */ - private final int poolNumber; - - // Utilities for CASing fields. Note that most of these - // are usually manually inlined by callers + volatile int blockedCount; /** - * Increments running count part of workerCounts. + * Counter for worker Thread names (unrelated to their poolIndex) */ - final void incrementRunningCount() { - int c; - do {} while (!UNSAFE.compareAndSwapInt(this, workerCountsOffset, - c = workerCounts, - c + ONE_RUNNING)); - } + private volatile int nextWorkerNumber; /** - * Tries to increment running count part of workerCounts. + * The index for the next created worker. Accessed under scanGuard. */ - final boolean tryIncrementRunningCount() { - int c; - return UNSAFE.compareAndSwapInt(this, workerCountsOffset, - c = workerCounts, - c + ONE_RUNNING); - } + private int nextWorkerIndex; /** - * Tries to decrement running count unless already zero. + * SeqLock and index masking for for updates to workers array. + * Locked when SG_UNIT is set. Unlocking clears bit by adding + * SG_UNIT. Staleness of read-only operations can be checked by + * comparing scanGuard to value before the reads. The low 16 bits + * (i.e, anding with SMASK) hold (the smallest power of two + * covering all worker indices, minus one, and is used to avoid + * dealing with large numbers of null slots when the workers array + * is overallocated. */ - final boolean tryDecrementRunningCount() { - int wc = workerCounts; - if ((wc & RUNNING_COUNT_MASK) == 0) - return false; - return UNSAFE.compareAndSwapInt(this, workerCountsOffset, - wc, wc - ONE_RUNNING); - } + volatile int scanGuard; - /** - * Forces decrement of encoded workerCounts, awaiting nonzero if - * (rarely) necessary when other count updates lag. - * - * @param dr -- either zero or ONE_RUNNING - * @param dt -- either zero or ONE_TOTAL - */ - private void decrementWorkerCounts(int dr, int dt) { - for (;;) { - int wc = workerCounts; - if ((wc & RUNNING_COUNT_MASK) - dr < 0 || - (wc >>> TOTAL_COUNT_SHIFT) - dt < 0) { - if ((runState & TERMINATED) != 0) - return; // lagging termination on a backout - Thread.yield(); - } - if (UNSAFE.compareAndSwapInt(this, workerCountsOffset, - wc, wc - (dr + dt))) - return; - } - } + private static final int SG_UNIT = 1 << 16; /** - * Tries decrementing active count; fails on contention. - * Called when workers cannot find tasks to run. + * The wakeup interval (in nanoseconds) for a worker waiting for a + * task when the pool is quiescent to instead try to shrink the + * number of workers. The exact value does not matter too + * much. It must be short enough to release resources during + * sustained periods of idleness, but not so short that threads + * are continually re-created. */ - final boolean tryDecrementActiveCount() { - int c; - return UNSAFE.compareAndSwapInt(this, runStateOffset, - c = runState, c - 1); - } + private static final long SHRINK_RATE = + 4L * 1000L * 1000L * 1000L; // 4 seconds /** - * Advances to at least the given level. Returns true if not - * already in at least the given level. + * Top-level loop for worker threads: On each step: if the + * previous step swept through all queues and found no tasks, or + * there are excess threads, then possibly blocks. Otherwise, + * scans for and, if found, executes a task. Returns when pool + * and/or worker terminate. + * + * @param w the worker */ - private boolean advanceRunLevel(int level) { - for (;;) { - int s = runState; - if ((s & level) != 0) - return false; - if (UNSAFE.compareAndSwapInt(this, runStateOffset, s, s | level)) - return true; + final void work(ForkJoinWorkerThread w) { + boolean swept = false; // true on empty scans + long c; + while (!w.terminate && (int)(c = ctl) >= 0) { + int a; // active count + if (!swept && (a = (int)(c >> AC_SHIFT)) <= 0) + swept = scan(w, a); + else if (tryAwaitWork(w, c)) + swept = false; } } - // workers array maintenance + // Signalling /** - * Records and returns a workers array index for new worker. + * Wakes up or creates a worker. */ - private int recordWorker(ForkJoinWorkerThread w) { - // Try using slot totalCount-1. If not available, scan and/or resize - int k = (workerCounts >>> TOTAL_COUNT_SHIFT) - 1; - final ReentrantLock lock = this.workerLock; - lock.lock(); - try { - ForkJoinWorkerThread[] ws = workers; - int n = ws.length; - if (k < 0 || k >= n || ws[k] != null) { - for (k = 0; k < n && ws[k] != null; ++k) - ; - if (k == n) - ws = workers = Arrays.copyOf(ws, n << 1); - } - ws[k] = w; - int c = eventCount; // advance event count to ensure visibility - UNSAFE.compareAndSwapInt(this, eventCountOffset, c, c+1); - } finally { - lock.unlock(); + final void signalWork() { + /* + * The while condition is true if: (there is are too few total + * workers OR there is at least one waiter) AND (there are too + * few active workers OR the pool is terminating). The value + * of e distinguishes the remaining cases: zero (no waiters) + * for create, negative if terminating (in which case do + * nothing), else release a waiter. The secondary checks for + * release (non-null array etc) can fail if the pool begins + * terminating after the test, and don't impose any added cost + * because JVMs must perform null and bounds checks anyway. + */ + long c; int e, u; + while ((((e = (int)(c = ctl)) | (u = (int)(c >>> 32))) & + (INT_SIGN|SHORT_SIGN)) == (INT_SIGN|SHORT_SIGN) && e >= 0) { + if (e > 0) { // release a waiting worker + int i; ForkJoinWorkerThread w; ForkJoinWorkerThread[] ws; + if ((ws = workers) == null || + (i = ~e & SMASK) >= ws.length || + (w = ws[i]) == null) + break; + long nc = (((long)(w.nextWait & E_MASK)) | + ((long)(u + UAC_UNIT) << 32)); + if (w.eventCount == e && + UNSAFE.compareAndSwapLong(this, ctlOffset, c, nc)) { + w.eventCount = (e + EC_UNIT) & E_MASK; + if (w.parked) + UNSAFE.unpark(w); + break; + } + } + else if (UNSAFE.compareAndSwapLong + (this, ctlOffset, c, + (long)(((u + UTC_UNIT) & UTC_MASK) | + ((u + UAC_UNIT) & UAC_MASK)) << 32)) { + addWorker(); + break; + } } - return k; } /** - * Nulls out record of worker in workers array. + * Variant of signalWork to help release waiters on rescans. + * Tries once to release a waiter if active count < 0. + * + * @return false if failed due to contention, else true */ - private void forgetWorker(ForkJoinWorkerThread w) { - int idx = w.poolIndex; - // Locking helps method recordWorker avoid unnecessary expansion - final ReentrantLock lock = this.workerLock; - lock.lock(); - try { - ForkJoinWorkerThread[] ws = workers; - if (idx >= 0 && idx < ws.length && ws[idx] == w) // verify - ws[idx] = null; - } finally { - lock.unlock(); + private boolean tryReleaseWaiter() { + long c; int e, i; ForkJoinWorkerThread w; ForkJoinWorkerThread[] ws; + if ((e = (int)(c = ctl)) > 0 && + (int)(c >> AC_SHIFT) < 0 && + (ws = workers) != null && + (i = ~e & SMASK) < ws.length && + (w = ws[i]) != null) { + long nc = ((long)(w.nextWait & E_MASK) | + ((c + AC_UNIT) & (AC_MASK|TC_MASK))); + if (w.eventCount != e || + !UNSAFE.compareAndSwapLong(this, ctlOffset, c, nc)) + return false; + w.eventCount = (e + EC_UNIT) & E_MASK; + if (w.parked) + UNSAFE.unpark(w); } + return true; } + // Scanning for tasks + /** - * Final callback from terminating worker. Removes record of - * worker from array, and adjusts counts. If pool is shutting - * down, tries to complete termination. + * Scans for and, if found, executes one task. Scans start at a + * random index of workers array, and randomly select the first + * (2*#workers)-1 probes, and then, if all empty, resort to 2 + * circular sweeps, which is necessary to check quiescence. and + * taking a submission only if no stealable tasks were found. The + * steal code inside the loop is a specialized form of + * ForkJoinWorkerThread.deqTask, followed bookkeeping to support + * helpJoinTask and signal propagation. The code for submission + * queues is almost identical. On each steal, the worker completes + * not only the task, but also all local tasks that this task may + * have generated. On detecting staleness or contention when + * trying to take a task, this method returns without finishing + * sweep, which allows global state rechecks before retry. * * @param w the worker + * @param a the number of active workers + * @return true if swept all queues without finding a task */ - final void workerTerminated(ForkJoinWorkerThread w) { - forgetWorker(w); - decrementWorkerCounts(w.isTrimmed() ? 0 : ONE_RUNNING, ONE_TOTAL); - while (w.stealCount != 0) // collect final count - tryAccumulateStealCount(w); - tryTerminate(false); - } - - // Waiting for and signalling events - - /** - * Releases workers blocked on a count not equal to current count. - * Normally called after precheck that eventWaiters isn't zero to - * avoid wasted array checks. Gives up upon a change in count or - * upon releasing four workers, letting others take over. - */ - private void releaseEventWaiters() { + private boolean scan(ForkJoinWorkerThread w, int a) { + int g = scanGuard; // mask 0 avoids useless scans if only one active + int m = parallelism == 1 - a? 0 : g & SMASK; ForkJoinWorkerThread[] ws = workers; - int n = ws.length; - long h = eventWaiters; - int ec = eventCount; - int releases = 4; - ForkJoinWorkerThread w; int id; - while ((id = (((int)h) & WAITER_ID_MASK) - 1) >= 0 && - (int)(h >>> EVENT_COUNT_SHIFT) != ec && - id < n && (w = ws[id]) != null) { - if (UNSAFE.compareAndSwapLong(this, eventWaitersOffset, - h, w.nextWaiter)) { - LockSupport.unpark(w); - if (--releases == 0) - break; + if (ws == null || ws.length <= m) // staleness check + return false; + for (int r = w.seed, k = r, j = -(m + m); j <= m + m; ++j) { + ForkJoinTask t; ForkJoinTask[] q; int b, i; + ForkJoinWorkerThread v = ws[k & m]; + if (v != null && (b = v.queueBase) != v.queueTop && + (q = v.queue) != null && (i = (q.length - 1) & b) >= 0) { + long u = (i << ASHIFT) + ABASE; + if ((t = q[i]) != null && v.queueBase == b && + UNSAFE.compareAndSwapObject(q, u, t, null)) { + int d = (v.queueBase = b + 1) - v.queueTop; + v.stealHint = w.poolIndex; + if (d != 0) + signalWork(); // propagate if nonempty + w.execTask(t); + } + r ^= r << 13; r ^= r >>> 17; w.seed = r ^ (r << 5); + return false; // store next seed } - if (eventCount != ec) - break; - h = eventWaiters; + else if (j < 0) { // xorshift + r ^= r << 13; r ^= r >>> 17; k = r ^= r << 5; + } + else + ++k; + } + if (scanGuard != g) // staleness check + return false; + else { // try to take submission + ForkJoinTask t; ForkJoinTask[] q; int b, i; + if ((b = queueBase) != queueTop && + (q = submissionQueue) != null && + (i = (q.length - 1) & b) >= 0) { + long u = (i << ASHIFT) + ABASE; + if ((t = q[i]) != null && queueBase == b && + UNSAFE.compareAndSwapObject(q, u, t, null)) { + queueBase = b + 1; + w.execTask(t); + } + return false; + } + return true; // all queues empty } } /** - * Tries to advance eventCount and releases waiters. Called only - * from workers. + * Tries to enqueue worker in wait queue and await change in + * worker's eventCount. Before blocking, rescans queues to avoid + * missed signals. If the pool is quiescent, possibly terminates + * worker upon exit. + * + * @param w the calling worker + * @param c the ctl value on entry + * @return true if waited or another thread was released upon enq */ - final void signalWork() { - int c; // try to increment event count -- CAS failure OK - UNSAFE.compareAndSwapInt(this, eventCountOffset, c = eventCount, c+1); - if (eventWaiters != 0L) - releaseEventWaiters(); - } - - /** - * Adds the given worker to event queue and blocks until - * terminating or event count advances from the given value - * - * @param w the calling worker thread - * @param ec the count - */ - private void eventSync(ForkJoinWorkerThread w, int ec) { - long nh = (((long)ec) << EVENT_COUNT_SHIFT) | ((long)(w.poolIndex+1)); - long h; - while ((runState < SHUTDOWN || !tryTerminate(false)) && - (((int)(h = eventWaiters) & WAITER_ID_MASK) == 0 || - (int)(h >>> EVENT_COUNT_SHIFT) == ec) && - eventCount == ec) { - if (UNSAFE.compareAndSwapLong(this, eventWaitersOffset, - w.nextWaiter = h, nh)) { - awaitEvent(w, ec); - break; + private boolean tryAwaitWork(ForkJoinWorkerThread w, long c) { + int v = w.eventCount; + w.nextWait = (int)c; // w's successor record + long nc = (long)(v & E_MASK) | ((c - AC_UNIT) & (AC_MASK|TC_MASK)); + if (ctl != c || !UNSAFE.compareAndSwapLong(this, ctlOffset, c, nc)) { + long d = ctl; // return true if lost to a deq, to force rescan + return (int)d != (int)c && ((d - c) & AC_MASK) >= 0L; + } + boolean rescanned = false; + for (int sc;;) { + if (w.eventCount != v) + return true; + if ((sc = w.stealCount) != 0) { + long s = stealCount; // accumulate stealCount + if (UNSAFE.compareAndSwapLong(this, stealCountOffset, s, s+sc)) + w.stealCount = 0; + } + else if (!rescanned) { + int g = scanGuard, m = g & SMASK; + ForkJoinWorkerThread[] ws = workers; + if (ws != null && m < ws.length) { + rescanned = true; + for (int i = 0; i <= m; ++i) { + ForkJoinWorkerThread u = ws[i]; + if (u != null) { + if (u.queueBase != u.queueTop && + !tryReleaseWaiter()) + rescanned = false; // contended + if (w.eventCount != v) + return true; + } + } + } + if (scanGuard != g || // stale + (queueBase != queueTop && !tryReleaseWaiter())) + rescanned = false; + if (!rescanned) + Thread.yield(); // reduce contention + else + Thread.interrupted(); // clear before park + } + else if (parallelism + (int)(ctl >> AC_SHIFT) == 0 && + blockedCount == 0 && quiescerCount == 0) + idleAwaitWork(w, v); // quiescent -- maybe shrink + else { + w.parked = true; // must recheck + if (w.eventCount != v) { + w.parked = false; + return true; + } + LockSupport.park(this); + rescanned = w.parked = false; } } } /** - * Blocks the given worker (that has already been entered as an - * event waiter) until terminating or event count advances from - * the given value. The oldest (first) waiter uses a timed wait to - * occasionally one-by-one shrink the number of workers (to a - * minimum of one) if the pool has not been used for extended - * periods. - * - * @param w the calling worker thread - * @param ec the count - */ - private void awaitEvent(ForkJoinWorkerThread w, int ec) { - while (eventCount == ec) { - if (tryAccumulateStealCount(w)) { // transfer while idle - boolean untimed = (w.nextWaiter != 0L || - (workerCounts & RUNNING_COUNT_MASK) <= 1); - long startTime = untimed ? 0 : System.nanoTime(); - Thread.interrupted(); // clear/ignore interrupt - if (w.isTerminating() || eventCount != ec) - break; // recheck after clear - if (untimed) - LockSupport.park(w); - else { - LockSupport.parkNanos(w, SHRINK_RATE_NANOS); - if (eventCount != ec || w.isTerminating()) - break; - if (System.nanoTime() - startTime >= SHRINK_RATE_NANOS) - tryShutdownUnusedWorker(ec); + * If pool is quiescent, checks for termination, and waits for + * event signal for up to SHRINK_RATE nanosecs. On timeout, if ctl + * has not changed, terminates the worker. Upon its termination + * (see deregisterWorker), it may wake up another worker to + * possibly repeat this process. + * + * @param w the calling worker + * @param v the eventCount w must wait until changed + */ + private void idleAwaitWork(ForkJoinWorkerThread w, int v) { + ForkJoinTask.helpExpungeStaleExceptions(); // help clean weak refs + if (shutdown) + tryTerminate(false); + long c = ctl; + long nc = (((c & (AC_MASK|TC_MASK)) + AC_UNIT) | + (long)(w.nextWait & E_MASK)); // ctl value to release w + if (w.eventCount == v && + parallelism + (int)(c >> AC_SHIFT) == 0 && + blockedCount == 0 && quiescerCount == 0) { + long startTime = System.nanoTime(); + Thread.interrupted(); + if (w.eventCount == v) { + w.parked = true; + if (w.eventCount == v) + LockSupport.parkNanos(this, SHRINK_RATE); + w.parked = false; + if (w.eventCount == v && ctl == c && + System.nanoTime() - startTime >= SHRINK_RATE && + UNSAFE.compareAndSwapLong(this, ctlOffset, c, nc)) { + w.terminate = true; + w.eventCount = ((int)c + EC_UNIT) & E_MASK; } } } } - // Maintaining parallelism + // Submissions /** - * Pushes worker onto the spare stack. + * Enqueues the given task in the submissionQueue. Same idea as + * ForkJoinWorkerThread.pushTask except for use of submissionLock. + * + * @param t the task */ - final void pushSpare(ForkJoinWorkerThread w) { - int ns = (++w.spareCount << SPARE_COUNT_SHIFT) | (w.poolIndex + 1); - do {} while (!UNSAFE.compareAndSwapInt(this, spareWaitersOffset, - w.nextSpare = spareWaiters,ns)); + private void addSubmission(ForkJoinTask t) { + final ReentrantLock lock = this.submissionLock; + lock.lock(); + try { + ForkJoinTask[] q; int s, m; + if ((q = submissionQueue) != null) { // ignore if queue removed + long u = (((s = queueTop) & (m = q.length-1)) << ASHIFT)+ABASE; + UNSAFE.putOrderedObject(q, u, t); + queueTop = s + 1; + if (s - queueBase == m) + growSubmissionQueue(); + } + } finally { + lock.unlock(); + } + signalWork(); } + // (pollSubmission is defined below with exported methods) + /** - * Tries (once) to resume a spare if the number of running - * threads is less than target. + * Creates or doubles submissionQueue array. + * Basically identical to ForkJoinWorkerThread version */ - private void tryResumeSpare() { - int sw, id; - ForkJoinWorkerThread[] ws = workers; - int n = ws.length; - ForkJoinWorkerThread w; - if ((sw = spareWaiters) != 0 && - (id = (sw & SPARE_ID_MASK) - 1) >= 0 && - id < n && (w = ws[id]) != null && - (runState >= TERMINATING || - (workerCounts & RUNNING_COUNT_MASK) < parallelism) && - spareWaiters == sw && - UNSAFE.compareAndSwapInt(this, spareWaitersOffset, - sw, w.nextSpare)) { - int c; // increment running count before resume - do {} while (!UNSAFE.compareAndSwapInt - (this, workerCountsOffset, - c = workerCounts, c + ONE_RUNNING)); - if (w.tryUnsuspend()) - LockSupport.unpark(w); - else // back out if w was shutdown - decrementWorkerCounts(ONE_RUNNING, 0); + private void growSubmissionQueue() { + ForkJoinTask[] oldQ = submissionQueue; + int size = oldQ != null ? oldQ.length << 1 : INITIAL_QUEUE_CAPACITY; + if (size > MAXIMUM_QUEUE_CAPACITY) + throw new RejectedExecutionException("Queue capacity exceeded"); + if (size < INITIAL_QUEUE_CAPACITY) + size = INITIAL_QUEUE_CAPACITY; + ForkJoinTask[] q = submissionQueue = new ForkJoinTask[size]; + int mask = size - 1; + int top = queueTop; + int oldMask; + if (oldQ != null && (oldMask = oldQ.length - 1) >= 0) { + for (int b = queueBase; b != top; ++b) { + long u = ((b & oldMask) << ASHIFT) + ABASE; + Object x = UNSAFE.getObjectVolatile(oldQ, u); + if (x != null && UNSAFE.compareAndSwapObject(oldQ, u, x, null)) + UNSAFE.putObjectVolatile + (q, ((b & mask) << ASHIFT) + ABASE, x); + } } } + // Blocking support + /** - * Tries to increase the number of running workers if below target - * parallelism: If a spare exists tries to resume it via - * tryResumeSpare. Otherwise, if not enough total workers or all - * existing workers are busy, adds a new worker. In all cases also - * helps wake up releasable workers waiting for work. - */ - private void helpMaintainParallelism() { - int pc = parallelism; - int wc, rs, tc; - while (((wc = workerCounts) & RUNNING_COUNT_MASK) < pc && - (rs = runState) < TERMINATING) { - if (spareWaiters != 0) - tryResumeSpare(); - else if ((tc = wc >>> TOTAL_COUNT_SHIFT) >= MAX_WORKERS || - (tc >= pc && (rs & ACTIVE_COUNT_MASK) != tc)) - break; // enough total - else if (runState == rs && workerCounts == wc && - UNSAFE.compareAndSwapInt(this, workerCountsOffset, wc, - wc + (ONE_RUNNING|ONE_TOTAL))) { - ForkJoinWorkerThread w = null; - Throwable fail = null; - try { - w = factory.newThread(this); - } catch (Throwable ex) { - fail = ex; + * Tries to increment blockedCount, decrement active count + * (sometimes implicitly) and possibly release or create a + * compensating worker in preparation for blocking. Fails + * on contention or termination. + * + * @return true if the caller can block, else should recheck and retry + */ + private boolean tryPreBlock() { + int b = blockedCount; + if (UNSAFE.compareAndSwapInt(this, blockedCountOffset, b, b + 1)) { + int pc = parallelism; + do { + ForkJoinWorkerThread[] ws; ForkJoinWorkerThread w; + int e, ac, tc, rc, i; + long c = ctl; + int u = (int)(c >>> 32); + if ((e = (int)c) < 0) { + // skip -- terminating } - if (w == null) { // null or exceptional factory return - decrementWorkerCounts(ONE_RUNNING, ONE_TOTAL); - tryTerminate(false); // handle failure during shutdown - // If originating from an external caller, - // propagate exception, else ignore - if (fail != null && runState < TERMINATING && - !(Thread.currentThread() instanceof - ForkJoinWorkerThread)) - UNSAFE.throwException(fail); - break; + else if ((ac = (u >> UAC_SHIFT)) <= 0 && e != 0 && + (ws = workers) != null && + (i = ~e & SMASK) < ws.length && + (w = ws[i]) != null) { + long nc = ((long)(w.nextWait & E_MASK) | + (c & (AC_MASK|TC_MASK))); + if (w.eventCount == e && + UNSAFE.compareAndSwapLong(this, ctlOffset, c, nc)) { + w.eventCount = (e + EC_UNIT) & E_MASK; + if (w.parked) + UNSAFE.unpark(w); + return true; // release an idle worker + } } - w.start(recordWorker(w), ueh); - if ((workerCounts >>> TOTAL_COUNT_SHIFT) >= pc) - break; // add at most one unless total below target - } + else if ((tc = (short)(u >>> UTC_SHIFT)) >= 0 && ac + pc > 1) { + long nc = ((c - AC_UNIT) & AC_MASK) | (c & ~AC_MASK); + if (UNSAFE.compareAndSwapLong(this, ctlOffset, c, nc)) + return true; // no compensation needed + } + else if (tc + pc < MAX_ID) { + long nc = ((c + TC_UNIT) & TC_MASK) | (c & ~TC_MASK); + if (UNSAFE.compareAndSwapLong(this, ctlOffset, c, nc)) { + addWorker(); + return true; // create a replacement + } + } + // try to back out on any failure and let caller retry + } while (!UNSAFE.compareAndSwapInt(this, blockedCountOffset, + b = blockedCount, b - 1)); } - if (eventWaiters != 0L) - releaseEventWaiters(); + return false; } /** - * Callback from the oldest waiter in awaitEvent waking up after a - * period of non-use. If all workers are idle, tries (once) to - * shutdown an event waiter or a spare, if one exists. Note that - * we don't need CAS or locks here because the method is called - * only from one thread occasionally waking (and even misfires are - * OK). Note that until the shutdown worker fully terminates, - * workerCounts will overestimate total count, which is tolerable. - * - * @param ec the event count waited on by caller (to abort - * attempt if count has since changed). + * Decrements blockedCount and increments active count */ - private void tryShutdownUnusedWorker(int ec) { - if (runState == 0 && eventCount == ec) { // only trigger if all idle - ForkJoinWorkerThread[] ws = workers; - int n = ws.length; - ForkJoinWorkerThread w = null; - boolean shutdown = false; - int sw; - long h; - if ((sw = spareWaiters) != 0) { // prefer killing spares - int id = (sw & SPARE_ID_MASK) - 1; - if (id >= 0 && id < n && (w = ws[id]) != null && - UNSAFE.compareAndSwapInt(this, spareWaitersOffset, - sw, w.nextSpare)) - shutdown = true; - } - else if ((h = eventWaiters) != 0L) { - long nh; - int id = (((int)h) & WAITER_ID_MASK) - 1; - if (id >= 0 && id < n && (w = ws[id]) != null && - (nh = w.nextWaiter) != 0L && // keep at least one worker - UNSAFE.compareAndSwapLong(this, eventWaitersOffset, h, nh)) - shutdown = true; - } - if (w != null && shutdown) { - w.shutdown(); - LockSupport.unpark(w); - } - } - releaseEventWaiters(); // in case of interference + private void postBlock() { + long c; + do {} while (!UNSAFE.compareAndSwapLong(this, ctlOffset, // no mask + c = ctl, c + AC_UNIT)); + int b; + do {} while(!UNSAFE.compareAndSwapInt(this, blockedCountOffset, + b = blockedCount, b - 1)); } /** - * Callback from workers invoked upon each top-level action (i.e., - * stealing a task or taking a submission and running it). - * Performs one or more of the following: - * - * 1. If the worker is active and either did not run a task - * or there are too many workers, try to set its active status - * to inactive and update activeCount. On contention, we may - * try again in this or a subsequent call. - * - * 2. If not enough total workers, help create some. - * - * 3. If there are too many running workers, suspend this worker - * (first forcing inactive if necessary). If it is not needed, - * it may be shutdown while suspended (via - * tryShutdownUnusedWorker). Otherwise, upon resume it - * rechecks running thread count and need for event sync. - * - * 4. If worker did not run a task, await the next task event via - * eventSync if necessary (first forcing inactivation), upon - * which the worker may be shutdown via - * tryShutdownUnusedWorker. Otherwise, help release any - * existing event waiters that are now releasable, + * Possibly blocks waiting for the given task to complete, or + * cancels the task if terminating. Fails to wait if contended. * - * @param w the worker - * @param ran true if worker ran a task since last call to this method + * @param joinMe the task */ - final void preStep(ForkJoinWorkerThread w, boolean ran) { - int wec = w.lastEventCount; - boolean active = w.active; - boolean inactivate = false; - int pc = parallelism; - while (w.runState == 0) { - int rs = runState; - if (rs >= TERMINATING) { // propagate shutdown - w.shutdown(); - break; - } - if ((inactivate || (active && (rs & ACTIVE_COUNT_MASK) >= pc)) && - UNSAFE.compareAndSwapInt(this, runStateOffset, rs, --rs)) { - inactivate = active = w.active = false; - if (rs == SHUTDOWN) { // all inactive and shut down - tryTerminate(false); - continue; - } - } - int wc = workerCounts; // try to suspend as spare - if ((wc & RUNNING_COUNT_MASK) > pc) { - if (!(inactivate |= active) && // must inactivate to suspend - workerCounts == wc && - UNSAFE.compareAndSwapInt(this, workerCountsOffset, - wc, wc - ONE_RUNNING)) - w.suspendAsSpare(); - } - else if ((wc >>> TOTAL_COUNT_SHIFT) < pc) - helpMaintainParallelism(); // not enough workers - else if (ran) - break; - else { - long h = eventWaiters; - int ec = eventCount; - if (h != 0L && (int)(h >>> EVENT_COUNT_SHIFT) != ec) - releaseEventWaiters(); // release others before waiting - else if (ec != wec) { - w.lastEventCount = ec; // no need to wait - break; - } - else if (!(inactivate |= active)) - eventSync(w, wec); // must inactivate before sync + final void tryAwaitJoin(ForkJoinTask joinMe) { + int s; + Thread.interrupted(); // clear interrupts before checking termination + if (joinMe.status >= 0) { + if (tryPreBlock()) { + joinMe.tryAwaitDone(0L); + postBlock(); } + if ((ctl & STOP_BIT) != 0L) + joinMe.cancelIgnoringExceptions(); } } /** - * Helps and/or blocks awaiting join of the given task. - * See above for explanation. + * Possibly blocks the given worker waiting for joinMe to + * complete or timeout * - * @param joinMe the task to join - * @param worker the current worker thread - * @param timed true if wait should time out - * @param nanos timeout value if timed + * @param joinMe the task + * @param millis the wait time for underlying Object.wait */ - final void awaitJoin(ForkJoinTask joinMe, ForkJoinWorkerThread worker, - boolean timed, long nanos) { - long startTime = timed ? System.nanoTime() : 0L; - int retries = 2 + (parallelism >> 2); // #helpJoins before blocking - boolean running = true; // false when count decremented + final void timedAwaitJoin(ForkJoinTask joinMe, long nanos) { while (joinMe.status >= 0) { - if (runState >= TERMINATING) { + Thread.interrupted(); + if ((ctl & STOP_BIT) != 0L) { joinMe.cancelIgnoringExceptions(); break; } - running = worker.helpJoinTask(joinMe, running); - if (joinMe.status < 0) - break; - if (retries > 0) { - --retries; - continue; - } - int wc = workerCounts; - if ((wc & RUNNING_COUNT_MASK) != 0) { - if (running) { - if (!UNSAFE.compareAndSwapInt(this, workerCountsOffset, - wc, wc - ONE_RUNNING)) - continue; - running = false; - } - long h = eventWaiters; - if (h != 0L && (int)(h >>> EVENT_COUNT_SHIFT) != eventCount) - releaseEventWaiters(); - if ((workerCounts & RUNNING_COUNT_MASK) != 0) { - long ms; int ns; - if (!timed) { - ms = JOIN_TIMEOUT_MILLIS; - ns = 0; - } - else { // at most JOIN_TIMEOUT_MILLIS per wait - long nt = nanos - (System.nanoTime() - startTime); - if (nt <= 0L) - break; - ms = nt / 1000000; - if (ms > JOIN_TIMEOUT_MILLIS) { - ms = JOIN_TIMEOUT_MILLIS; - ns = 0; - } - else - ns = (int) (nt % 1000000); + if (tryPreBlock()) { + long last = System.nanoTime(); + while (joinMe.status >= 0) { + long millis = TimeUnit.NANOSECONDS.toMillis(nanos); + if (millis <= 0) + break; + joinMe.tryAwaitDone(millis); + if (joinMe.status < 0) + break; + if ((ctl & STOP_BIT) != 0L) { + joinMe.cancelIgnoringExceptions(); + break; } - joinMe.internalAwaitDone(ms, ns); + long now = System.nanoTime(); + nanos -= now - last; + last = now; } - if (joinMe.status < 0) - break; + postBlock(); + break; } - helpMaintainParallelism(); - } - if (!running) { - int c; - do {} while (!UNSAFE.compareAndSwapInt - (this, workerCountsOffset, - c = workerCounts, c + ONE_RUNNING)); } } /** - * Same idea as awaitJoin, but no helping, retries, or timeouts. + * If necessary, compensates for blocker, and blocks */ - final void awaitBlocker(ManagedBlocker blocker) + private void awaitBlocker(ManagedBlocker blocker) throws InterruptedException { while (!blocker.isReleasable()) { - int wc = workerCounts; - if ((wc & RUNNING_COUNT_MASK) == 0) - helpMaintainParallelism(); - else if (UNSAFE.compareAndSwapInt(this, workerCountsOffset, - wc, wc - ONE_RUNNING)) { + if (tryPreBlock()) { try { - while (!blocker.isReleasable()) { - long h = eventWaiters; - if (h != 0L && - (int)(h >>> EVENT_COUNT_SHIFT) != eventCount) - releaseEventWaiters(); - else if ((workerCounts & RUNNING_COUNT_MASK) == 0 && - runState < TERMINATING) - helpMaintainParallelism(); - else if (blocker.block()) - break; - } + do {} while (!blocker.isReleasable() && !blocker.block()); } finally { - int c; - do {} while (!UNSAFE.compareAndSwapInt - (this, workerCountsOffset, - c = workerCounts, c + ONE_RUNNING)); + postBlock(); } break; } } } + // Creating, registering and deregistring workers + + /** + * Tries to create and start a worker; minimally rolls back counts + * on failure. + */ + private void addWorker() { + Throwable ex = null; + ForkJoinWorkerThread t = null; + try { + t = factory.newThread(this); + } catch (Throwable e) { + ex = e; + } + if (t == null) { // null or exceptional factory return + long c; // adjust counts + do {} while (!UNSAFE.compareAndSwapLong + (this, ctlOffset, c = ctl, + (((c - AC_UNIT) & AC_MASK) | + ((c - TC_UNIT) & TC_MASK) | + (c & ~(AC_MASK|TC_MASK))))); + // Propagate exception if originating from an external caller + if (!tryTerminate(false) && ex != null && + !(Thread.currentThread() instanceof ForkJoinWorkerThread)) + UNSAFE.throwException(ex); + } + else + t.start(); + } + + /** + * Callback from ForkJoinWorkerThread constructor to assign a + * public name + */ + final String nextWorkerName() { + for (int n;;) { + if (UNSAFE.compareAndSwapInt(this, nextWorkerNumberOffset, + n = nextWorkerNumber, ++n)) + return workerNamePrefix + n; + } + } + + /** + * Callback from ForkJoinWorkerThread constructor to + * determine its poolIndex and record in workers array. + * + * @param w the worker + * @return the worker's pool index + */ + final int registerWorker(ForkJoinWorkerThread w) { + /* + * In the typical case, a new worker acquires the lock, uses + * next available index and returns quickly. Since we should + * not block callers (ultimately from signalWork or + * tryPreBlock) waiting for the lock needed to do this, we + * instead help release other workers while waiting for the + * lock. + */ + for (int g;;) { + ForkJoinWorkerThread[] ws; + if (((g = scanGuard) & SG_UNIT) == 0 && + UNSAFE.compareAndSwapInt(this, scanGuardOffset, + g, g | SG_UNIT)) { + int k = nextWorkerIndex; + try { + if ((ws = workers) != null) { // ignore on shutdown + int n = ws.length; + if (k < 0 || k >= n || ws[k] != null) { + for (k = 0; k < n && ws[k] != null; ++k) + ; + if (k == n) + ws = workers = Arrays.copyOf(ws, n << 1); + } + ws[k] = w; + nextWorkerIndex = k + 1; + int m = g & SMASK; + g = k >= m? ((m << 1) + 1) & SMASK : g + (SG_UNIT<<1); + } + } finally { + scanGuard = g; + } + return k; + } + else if ((ws = workers) != null) { // help release others + for (ForkJoinWorkerThread u : ws) { + if (u != null && u.queueBase != u.queueTop) { + if (tryReleaseWaiter()) + break; + } + } + } + } + } + + /** + * Final callback from terminating worker. Removes record of + * worker from array, and adjusts counts. If pool is shutting + * down, tries to complete termination. + * + * @param w the worker + */ + final void deregisterWorker(ForkJoinWorkerThread w, Throwable ex) { + int idx = w.poolIndex; + int sc = w.stealCount; + int steps = 0; + // Remove from array, adjust worker counts and collect steal count. + // We can intermix failed removes or adjusts with steal updates + do { + long s, c; + int g; + if (steps == 0 && ((g = scanGuard) & SG_UNIT) == 0 && + UNSAFE.compareAndSwapInt(this, scanGuardOffset, + g, g |= SG_UNIT)) { + ForkJoinWorkerThread[] ws = workers; + if (ws != null && idx >= 0 && + idx < ws.length && ws[idx] == w) + ws[idx] = null; // verify + nextWorkerIndex = idx; + scanGuard = g + SG_UNIT; + steps = 1; + } + if (steps == 1 && + UNSAFE.compareAndSwapLong(this, ctlOffset, c = ctl, + (((c - AC_UNIT) & AC_MASK) | + ((c - TC_UNIT) & TC_MASK) | + (c & ~(AC_MASK|TC_MASK))))) + steps = 2; + if (sc != 0 && + UNSAFE.compareAndSwapLong(this, stealCountOffset, + s = stealCount, s + sc)) + sc = 0; + } while (steps != 2 || sc != 0); + if (!tryTerminate(false)) { + if (ex != null) // possibly replace if died abnormally + signalWork(); + else + tryReleaseWaiter(); + } + } + + // Shutdown and termination + /** * Possibly initiates and/or completes termination. * @@ -1132,97 +1200,128 @@ public class ForkJoinPool extends Abstra * @return true if now terminating or terminated */ private boolean tryTerminate(boolean now) { - if (now) - advanceRunLevel(SHUTDOWN); // ensure at least SHUTDOWN - else if (runState < SHUTDOWN || - !submissionQueue.isEmpty() || - (runState & ACTIVE_COUNT_MASK) != 0) - return false; - - if (advanceRunLevel(TERMINATING)) - startTerminating(); - - // Finish now if all threads terminated; else in some subsequent call - if ((workerCounts >>> TOTAL_COUNT_SHIFT) == 0) { - advanceRunLevel(TERMINATED); - termination.forceTermination(); + long c; + while (((c = ctl) & STOP_BIT) == 0) { + if (!now) { + if ((int)(c >> AC_SHIFT) != -parallelism) + return false; + if (!shutdown || blockedCount != 0 || quiescerCount != 0 || + queueTop - queueBase > 0) { + if (ctl == c) // staleness check + return false; + continue; + } + } + if (UNSAFE.compareAndSwapLong(this, ctlOffset, c, c | STOP_BIT)) + startTerminating(); + } + if ((short)(c >>> TC_SHIFT) == -parallelism) { + submissionLock.lock(); + termination.signalAll(); + submissionLock.unlock(); } return true; } /** - * Actions on transition to TERMINATING - * - * Runs up to four passes through workers: (0) shutting down each - * (without waking up if parked) to quickly spread notifications - * without unnecessary bouncing around event queues etc (1) wake - * up and help cancel tasks (2) interrupt (3) mop up races with - * interrupted workers + * Runs up to three passes through workers: (0) Setting + * termination status for each worker, followed by wakeups up + * queued workers (1) helping cancel tasks (2) interrupting + * lagging threads (likely in external tasks, but possibly also + * blocked in joins). Each pass repeats previous steps because of + * potential lagging thread creation. */ private void startTerminating() { cancelSubmissions(); - for (int passes = 0; passes < 4 && workerCounts != 0; ++passes) { - int c; // advance event count - UNSAFE.compareAndSwapInt(this, eventCountOffset, - c = eventCount, c+1); - eventWaiters = 0L; // clobber lists - spareWaiters = 0; - for (ForkJoinWorkerThread w : workers) { - if (w != null) { - w.shutdown(); - if (passes > 0 && !w.isTerminated()) { - w.cancelTasks(); - LockSupport.unpark(w); - if (passes > 1 && !w.isInterrupted()) { - try { - w.interrupt(); - } catch (SecurityException ignore) { + for (int pass = 0; pass < 3; ++pass) { + ForkJoinWorkerThread[] ws = workers; + if (ws != null) { + for (ForkJoinWorkerThread w : ws) { + if (w != null) { + w.terminate = true; + if (pass > 0) { + w.cancelTasks(); + if (pass > 1 && !w.isInterrupted()) { + try { + w.interrupt(); + } catch (SecurityException ignore) { + } } } } } + terminateWaiters(); } } } /** - * Clears out and cancels submissions, ignoring exceptions. + * Polls and cancels all submissions. Called only during termination. */ private void cancelSubmissions() { - ForkJoinTask task; - while ((task = submissionQueue.poll()) != null) { - try { - task.cancel(false); - } catch (Throwable ignore) { + while (queueBase != queueTop) { + ForkJoinTask task = pollSubmission(); + if (task != null) { + try { + task.cancel(false); + } catch (Throwable ignore) { + } } } } - // misc support for ForkJoinWorkerThread + /** + * Tries to set the termination status of waiting workers, and + * then wake them up (after which they will terminate). + */ + private void terminateWaiters() { + ForkJoinWorkerThread[] ws = workers; + if (ws != null) { + ForkJoinWorkerThread w; long c; int i, e; + int n = ws.length; + while ((i = ~(e = (int)(c = ctl)) & SMASK) < n && + (w = ws[i]) != null && w.eventCount == (e & E_MASK)) { + if (UNSAFE.compareAndSwapLong(this, ctlOffset, c, + (long)(w.nextWait & E_MASK) | + ((c + AC_UNIT) & AC_MASK) | + (c & (TC_MASK|STOP_BIT)))) { + w.terminate = true; + w.eventCount = e + EC_UNIT; + if (w.parked) + UNSAFE.unpark(w); + } + } + } + } + + // misc ForkJoinWorkerThread support /** - * Returns pool number. + * Increment or decrement quiescerCount. Needed only to prevent + * triggering shutdown if a worker is transiently inactive while + * checking quiescence. + * + * @param delta 1 for increment, -1 for decrement */ - final int getPoolNumber() { - return poolNumber; + final void addQuiescerCount(int delta) { + int c; + do {} while(!UNSAFE.compareAndSwapInt(this, quiescerCountOffset, + c = quiescerCount, c + delta)); } /** - * Tries to accumulate steal count from a worker, clearing - * the worker's value if successful. + * Directly increment or decrement active count without + * queuing. This method is used to transiently assert inactivation + * while checking quiescence. * - * @return true if worker steal count now zero + * @param delta 1 for increment, -1 for decrement */ - final boolean tryAccumulateStealCount(ForkJoinWorkerThread w) { - int sc = w.stealCount; - long c = stealCount; - // CAS even if zero, for fence effects - if (UNSAFE.compareAndSwapLong(this, stealCountOffset, c, c + sc)) { - if (sc != 0) - w.stealCount = 0; - return true; - } - return sc == 0; + final void addActiveCount(int delta) { + long d = delta < 0 ? -AC_UNIT : AC_UNIT; + long c; + do {} while (!UNSAFE.compareAndSwapLong(this, ctlOffset, c = ctl, + ((c + d) & AC_MASK) | + (c & ~AC_MASK))); } /** @@ -1230,16 +1329,17 @@ public class ForkJoinPool extends Abstra * active thread. */ final int idlePerActive() { - int pc = parallelism; // use parallelism, not rc - int ac = runState; // no mask -- artificially boosts during shutdown - // Use exact results for small values, saturate past 4 - return ((pc <= ac) ? 0 : - (pc >>> 1 <= ac) ? 1 : - (pc >>> 2 <= ac) ? 3 : - pc >>> 3); + // Approximate at powers of two for small values, saturate past 4 + int p = parallelism; + int a = p + (int)(ctl >> AC_SHIFT); + return (a > (p >>>= 1) ? 0 : + a > (p >>>= 1) ? 1 : + a > (p >>>= 1) ? 2 : + a > (p >>>= 1) ? 4 : + 8); } - // Public and protected methods + // Exported methods // Constructors @@ -1308,49 +1408,42 @@ public class ForkJoinPool extends Abstra checkPermission(); if (factory == null) throw new NullPointerException(); - if (parallelism <= 0 || parallelism > MAX_WORKERS) + if (parallelism <= 0 || parallelism > MAX_ID) throw new IllegalArgumentException(); this.parallelism = parallelism; this.factory = factory; this.ueh = handler; this.locallyFifo = asyncMode; - int arraySize = initialArraySizeFor(parallelism); - this.workers = new ForkJoinWorkerThread[arraySize]; - this.submissionQueue = new LinkedTransferQueue>(); - this.workerLock = new ReentrantLock(); - this.termination = new Phaser(1); - this.poolNumber = poolNumberGenerator.incrementAndGet(); - } - - /** - * Returns initial power of two size for workers array. - * @param pc the initial parallelism level - */ - private static int initialArraySizeFor(int pc) { - // If possible, initially allocate enough space for one spare - int size = pc < MAX_WORKERS ? pc + 1 : MAX_WORKERS; - // See Hackers Delight, sec 3.2. We know MAX_WORKERS < (1 >>> 16) - size |= size >>> 1; - size |= size >>> 2; - size |= size >>> 4; - size |= size >>> 8; - return size + 1; + long np = (long)(-parallelism); // offset ctl counts + this.ctl = ((np << AC_SHIFT) & AC_MASK) | ((np << TC_SHIFT) & TC_MASK); + this.submissionQueue = new ForkJoinTask[INITIAL_QUEUE_CAPACITY]; + // initialize workers array with room for 2*parallelism if possible + int n = parallelism << 1; + if (n >= MAX_ID) + n = MAX_ID; + else { // See Hackers Delight, sec 3.2, where n < (1 << 16) + n |= n >>> 1; n |= n >>> 2; n |= n >>> 4; n |= n >>> 8; + } + workers = new ForkJoinWorkerThread[n + 1]; + this.submissionLock = new ReentrantLock(); + this.termination = submissionLock.newCondition(); + StringBuilder sb = new StringBuilder("ForkJoinPool-"); + sb.append(poolNumberGenerator.incrementAndGet()); + sb.append("-worker-"); + this.workerNamePrefix = sb.toString(); } // Execution methods /** - * Submits task and creates, starts, or resumes some workers if necessary - */ - private void doSubmit(ForkJoinTask task) { - submissionQueue.offer(task); - int c; // try to increment event count -- CAS failure OK - UNSAFE.compareAndSwapInt(this, eventCountOffset, c = eventCount, c+1); - helpMaintainParallelism(); - } - - /** * Performs the given task, returning its result upon completion. + * If the computation encounters an unchecked Exception or Error, + * it is rethrown as the outcome of this invocation. Rethrown + * exceptions behave in the same way as regular exceptions, but, + * when possible, contain stack traces (as displayed for example + * using {@code ex.printStackTrace()}) of both the current thread + * as well as the thread actually encountering the exception; + * minimally only the latter. * * @param task the task * @return the task's result @@ -1359,16 +1452,16 @@ public class ForkJoinPool extends Abstra * scheduled for execution */ public T invoke(ForkJoinTask task) { + Thread t = Thread.currentThread(); if (task == null) throw new NullPointerException(); - if (runState >= SHUTDOWN) + if (shutdown) throw new RejectedExecutionException(); - Thread t = Thread.currentThread(); if ((t instanceof ForkJoinWorkerThread) && ((ForkJoinWorkerThread)t).pool == this) return task.invoke(); // bypass submit if in same pool else { - doSubmit(task); + addSubmission(task); return task.join(); } } @@ -1378,14 +1471,15 @@ public class ForkJoinPool extends Abstra * computation in the current pool, else submits as external task. */ private void forkOrSubmit(ForkJoinTask task) { - if (runState >= SHUTDOWN) - throw new RejectedExecutionException(); + ForkJoinWorkerThread w; Thread t = Thread.currentThread(); + if (shutdown) + throw new RejectedExecutionException(); if ((t instanceof ForkJoinWorkerThread) && - ((ForkJoinWorkerThread)t).pool == this) - task.fork(); + (w = (ForkJoinWorkerThread)t).pool == this) + w.pushTask(task); else - doSubmit(task); + addSubmission(task); } /** @@ -1542,7 +1636,7 @@ public class ForkJoinPool extends Abstra * @return the number of worker threads */ public int getPoolSize() { - return workerCounts >>> TOTAL_COUNT_SHIFT; + return parallelism + (short)(ctl >>> TC_SHIFT); } /** @@ -1564,7 +1658,8 @@ public class ForkJoinPool extends Abstra * @return the number of worker threads */ public int getRunningThreadCount() { - return workerCounts & RUNNING_COUNT_MASK; + int r = parallelism + (int)(ctl >> AC_SHIFT); + return r <= 0? 0 : r; // suppress momentarily negative values } /** @@ -1575,7 +1670,8 @@ public class ForkJoinPool extends Abstra * @return the number of active threads */ public int getActiveThreadCount() { - return runState & ACTIVE_COUNT_MASK; + int r = parallelism + (int)(ctl >> AC_SHIFT) + blockedCount; + return r <= 0? 0 : r; // suppress momentarily negative values } /** @@ -1590,7 +1686,7 @@ public class ForkJoinPool extends Abstra * @return {@code true} if all threads are currently idle */ public boolean isQuiescent() { - return (runState & ACTIVE_COUNT_MASK) == 0; + return parallelism + (int)(ctl >> AC_SHIFT) + blockedCount == 0; } /** @@ -1620,21 +1716,25 @@ public class ForkJoinPool extends Abstra */ public long getQueuedTaskCount() { long count = 0; - for (ForkJoinWorkerThread w : workers) - if (w != null) - count += w.getQueueSize(); + ForkJoinWorkerThread[] ws; + if ((short)(ctl >>> TC_SHIFT) > -parallelism && + (ws = workers) != null) { + for (ForkJoinWorkerThread w : ws) + if (w != null) + count -= w.queueBase - w.queueTop; // must read base first + } return count; } /** * Returns an estimate of the number of tasks submitted to this - * pool that have not yet begun executing. This method takes time - * proportional to the number of submissions. + * pool that have not yet begun executing. This meThod may take + * time proportional to the number of submissions. * * @return the number of queued submissions */ public int getQueuedSubmissionCount() { - return submissionQueue.size(); + return -queueBase + queueTop; } /** @@ -1644,7 +1744,7 @@ public class ForkJoinPool extends Abstra * @return {@code true} if there are any queued submissions */ public boolean hasQueuedSubmissions() { - return !submissionQueue.isEmpty(); + return queueBase != queueTop; } /** @@ -1655,7 +1755,19 @@ public class ForkJoinPool extends Abstra * @return the next submission, or {@code null} if none */ protected ForkJoinTask pollSubmission() { - return submissionQueue.poll(); + ForkJoinTask t; ForkJoinTask[] q; int b, i; + while ((b = queueBase) != queueTop && + (q = submissionQueue) != null && + (i = (q.length - 1) & b) >= 0) { + long u = (i << ASHIFT) + ABASE; + if ((t = q[i]) != null && + queueBase == b && + UNSAFE.compareAndSwapObject(q, u, t, null)) { + queueBase = b + 1; + return t; + } + } + return null; } /** @@ -1676,10 +1788,21 @@ public class ForkJoinPool extends Abstra * @return the number of elements transferred */ protected int drainTasksTo(Collection> c) { - int count = submissionQueue.drainTo(c); - for (ForkJoinWorkerThread w : workers) - if (w != null) - count += w.drainTasksTo(c); + int count = 0; + while (queueBase != queueTop) { + ForkJoinTask t = pollSubmission(); + if (t != null) { + c.add(t); + ++count; + } + } + ForkJoinWorkerThread[] ws; + if ((short)(ctl >>> TC_SHIFT) > -parallelism && + (ws = workers) != null) { + for (ForkJoinWorkerThread w : ws) + if (w != null) + count += w.drainTasksTo(c); + } return count; } @@ -1694,14 +1817,20 @@ public class ForkJoinPool extends Abstra long st = getStealCount(); long qt = getQueuedTaskCount(); long qs = getQueuedSubmissionCount(); - int wc = workerCounts; - int tc = wc >>> TOTAL_COUNT_SHIFT; - int rc = wc & RUNNING_COUNT_MASK; int pc = parallelism; - int rs = runState; - int ac = rs & ACTIVE_COUNT_MASK; + long c = ctl; + int tc = pc + (short)(c >>> TC_SHIFT); + int rc = pc + (int)(c >> AC_SHIFT); + if (rc < 0) // ignore transient negative + rc = 0; + int ac = rc + blockedCount; + String level; + if ((c & STOP_BIT) != 0) + level = (tc == 0)? "Terminated" : "Terminating"; + else + level = shutdown? "Shutting down" : "Running"; return super.toString() + - "[" + runLevelToString(rs) + + "[" + level + ", parallelism = " + pc + ", size = " + tc + ", active = " + ac + @@ -1712,13 +1841,6 @@ public class ForkJoinPool extends Abstra "]"; } - private static String runLevelToString(int s) { - return ((s & TERMINATED) != 0 ? "Terminated" : - ((s & TERMINATING) != 0 ? "Terminating" : - ((s & SHUTDOWN) != 0 ? "Shutting down" : - "Running"))); - } - /** * Initiates an orderly shutdown in which previously submitted * tasks are executed, but no new tasks will be accepted. @@ -1733,7 +1855,7 @@ public class ForkJoinPool extends Abstra */ public void shutdown() { checkPermission(); - advanceRunLevel(SHUTDOWN); + shutdown = true; tryTerminate(false); } @@ -1755,6 +1877,7 @@ public class ForkJoinPool extends Abstra */ public List shutdownNow() { checkPermission(); + shutdown = true; tryTerminate(true); return Collections.emptyList(); } @@ -1765,7 +1888,9 @@ public class ForkJoinPool extends Abstra * @return {@code true} if all tasks have completed following shut down */ public boolean isTerminated() { - return runState >= TERMINATED; + long c = ctl; + return ((c & STOP_BIT) != 0L && + (short)(c >>> TC_SHIFT) == -parallelism); } /** @@ -1782,14 +1907,16 @@ public class ForkJoinPool extends Abstra * @return {@code true} if terminating but not yet terminated */ public boolean isTerminating() { - return (runState & (TERMINATING|TERMINATED)) == TERMINATING; + long c = ctl; + return ((c & STOP_BIT) != 0L && + (short)(c >>> TC_SHIFT) != -parallelism); } /** * Returns true if terminating or terminated. Used by ForkJoinWorkerThread. */ final boolean isAtLeastTerminating() { - return runState >= TERMINATING; + return (ctl & STOP_BIT) != 0L; } /** @@ -1798,7 +1925,7 @@ public class ForkJoinPool extends Abstra * @return {@code true} if this pool has been shut down */ public boolean isShutdown() { - return runState >= SHUTDOWN; + return shutdown; } /** @@ -1814,12 +1941,20 @@ public class ForkJoinPool extends Abstra */ public boolean awaitTermination(long timeout, TimeUnit unit) throws InterruptedException { + long nanos = unit.toNanos(timeout); + final ReentrantLock lock = this.submissionLock; + lock.lock(); try { - termination.awaitAdvanceInterruptibly(0, timeout, unit); - } catch (TimeoutException ex) { - return false; + for (;;) { + if (isTerminated()) + return true; + if (nanos <= 0) + return false; + nanos = termination.awaitNanos(nanos); + } + } finally { + lock.unlock(); } - return true; } /** @@ -1938,30 +2073,47 @@ public class ForkJoinPool extends Abstra } // Unsafe mechanics - - private static final sun.misc.Unsafe UNSAFE = getUnsafe(); - private static final long workerCountsOffset = - objectFieldOffset("workerCounts", ForkJoinPool.class); - private static final long runStateOffset = - objectFieldOffset("runState", ForkJoinPool.class); - private static final long eventCountOffset = - objectFieldOffset("eventCount", ForkJoinPool.class); - private static final long eventWaitersOffset = - objectFieldOffset("eventWaiters", ForkJoinPool.class); - private static final long stealCountOffset = - objectFieldOffset("stealCount", ForkJoinPool.class); - private static final long spareWaitersOffset = - objectFieldOffset("spareWaiters", ForkJoinPool.class); - - private static long objectFieldOffset(String field, Class klazz) { + private static final sun.misc.Unsafe UNSAFE; + private static final long ctlOffset; + private static final long stealCountOffset; + private static final long blockedCountOffset; + private static final long quiescerCountOffset; + private static final long scanGuardOffset; + private static final long nextWorkerNumberOffset; + private static final long ABASE; + private static final int ASHIFT; + + static { + poolNumberGenerator = new AtomicInteger(); + workerSeedGenerator = new Random(); + modifyThreadPermission = new RuntimePermission("modifyThread"); + defaultForkJoinWorkerThreadFactory = + new DefaultForkJoinWorkerThreadFactory(); + int s; try { - return UNSAFE.objectFieldOffset(klazz.getDeclaredField(field)); - } catch (NoSuchFieldException e) { - // Convert Exception to corresponding Error - NoSuchFieldError error = new NoSuchFieldError(field); - error.initCause(e); - throw error; - } + UNSAFE = getUnsafe(); + Class k = ForkJoinPool.class; + ctlOffset = UNSAFE.objectFieldOffset + (k.getDeclaredField("ctl")); + stealCountOffset = UNSAFE.objectFieldOffset + (k.getDeclaredField("stealCount")); + blockedCountOffset = UNSAFE.objectFieldOffset + (k.getDeclaredField("blockedCount")); + quiescerCountOffset = UNSAFE.objectFieldOffset + (k.getDeclaredField("quiescerCount")); + scanGuardOffset = UNSAFE.objectFieldOffset + (k.getDeclaredField("scanGuard")); + nextWorkerNumberOffset = UNSAFE.objectFieldOffset + (k.getDeclaredField("nextWorkerNumber")); + Class a = ForkJoinTask[].class; + ABASE = UNSAFE.arrayBaseOffset(a); + s = UNSAFE.arrayIndexScale(a); + } catch (Exception e) { + throw new Error(e); + } + if ((s & (s-1)) != 0) + throw new Error("data type scale not a power of two"); + ASHIFT = 31 - Integer.numberOfLeadingZeros(s); } /**