bslmt_fastpostsemaphoreimpl.h

Go to the documentation of this file.

/// @file bslmt_fastpostsemaphoreimpl.h
///
/// The content of this file has been pre-processed for Doxygen.
///
 
 
// bslmt_fastpostsemaphoreimpl.h                                      -*-C++-*-
 
#ifndef INCLUDED_BSLMT_FASTPOSTSEMAPHOREIMPL
#define INCLUDED_BSLMT_FASTPOSTSEMAPHOREIMPL
 
#include <bsls_ident.h>
BSLS_IDENT("$Id: $")
 
/// @defgroup bslmt_fastpostsemaphoreimpl bslmt_fastpostsemaphoreimpl
/// @brief  Provide a testable semaphore class optimizing `post`.
/// @addtogroup bsl
/// @{
/// @addtogroup bslmt
/// @{
/// @addtogroup bslmt_fastpostsemaphoreimpl
/// @{
///
/// <h1> Outline </h1>
/// * <a href="#bslmt_fastpostsemaphoreimpl-purpose"> Purpose</a>
/// * <a href="#bslmt_fastpostsemaphoreimpl-classes"> Classes </a>
/// * <a href="#bslmt_fastpostsemaphoreimpl-description"> Description </a>
///   * <a href="#bslmt_fastpostsemaphoreimpl-supported-clock-types"> Supported Clock-Types </a>
///   * <a href="#bslmt_fastpostsemaphoreimpl-usage"> Usage </a>
///
/// # Purpose {#bslmt_fastpostsemaphoreimpl-purpose}
/// Provide a testable semaphore class optimizing `post`.
///
/// # Classes {#bslmt_fastpostsemaphoreimpl-classes}
///
/// -  bslmt::FastPostSemaphoreImpl: testable semaphore class optimizing `post`
///
/// @see  bslmt_fastpostsemaphore, bslmt_semaphore
///
/// # Description {#bslmt_fastpostsemaphoreimpl-description}
/// This component defines a testable semaphore,
/// `bslmt::FastPostSemaphoreImpl`, with the `post` operation being optimized at
/// the potential expense of other operations.  In particular,
/// `bslmt::FastPostSemaphoreImpl` is an efficient synchronization primitive
/// that enables sharing of a counted number of resources.
/// `bslmt::FastPostSemaphoreImpl` supports the methods `timedWait`, `enable`,
/// and `disable` in addition to the standard semaphore methods.
///
/// Commonly, during periods of time when the protected resource is scarce (the
/// semaphore count is frequently zero) and threads are frequently blocked on
/// wait methods, pessimizing the performance of the threads that block will
/// have little effect on overall performance.  In this case, optimizing `post`
/// *may* be a performance improvement.  Note that when the resource is
/// plentiful, there are no blocked threads and we expect the differences
/// between semaphore implementations to be trivial.
///
/// ## Supported Clock-Types {#bslmt_fastpostsemaphoreimpl-supported-clock-types}
///
///
/// `bsls::SystemClockType` supplies the enumeration indicating the system clock
/// on which timeouts supplied to other methods should be based.  If the clock
/// type indicated at construction is `bsls::SystemClockType::e_REALTIME`, the
/// `absTime` argument passed to the `timedWait` method should be expressed as
/// an *absolute* offset since 00:00:00 UTC, January 1, 1970 (which matches the
/// epoch used in `bsls::SystemTime::now(bsls::SystemClockType::e_REALTIME)`.
/// If the clock type indicated at construction is
/// `bsls::SystemClockType::e_MONOTONIC`, the `absTime` argument passed to the
/// `timedWait` method should be expressed as an *absolute* offset since the
/// epoch of this clock (which matches the epoch used in
/// `bsls::SystemTime::now(bsls::SystemClockType::e_MONOTONIC)`.
///
/// ## Usage {#bslmt_fastpostsemaphoreimpl-usage}
///
///
/// There is no usage example for this component since it is not meant for
/// direct client use.
/// @}
/** @} */
/** @} */
 
/** @addtogroup bsl
 * @{
 */
/** @addtogroup bslmt
 * @{
 */
/** @addtogroup bslmt_fastpostsemaphoreimpl
 * @{
 */
 
#include <bslscm_version.h>
 
#include <bslmt_lockguard.h>
 
#include <bsls_review.h>
#include <bsls_systemclocktype.h>
#include <bsls_timeinterval.h>
#include <bsls_types.h>
 
#include <bsl_climits.h>
 
 
namespace bslmt {
 
                // =========================================
                // class FastPostSemaphoreImplWorkaroundUtil
                // =========================================
 
/// This class provides utility functions for workarounds to system level
/// issues for `FastPostSemaphoreImpl`.
///
/// See @ref bslmt_fastpostsemaphoreimpl 
class FastPostSemaphoreImplWorkaroundUtil {
 
  private:
    // CLASS DATA
    static bool s_postAlwaysSignals;
 
  public:
    // PUBLIC CLASS METHODS
 
    /// Remove the mitigation of `post` always signalling the condition
    /// variable.  Note this mitigation was introduced as a work around for
    /// a lost signal bug in the underlying implementation of condition
    /// variable (e.g.,
    /// https://sourceware.org/bugzilla/show_bug.cgi?id=25847).  The
    /// availability of the pthread correction is tracked in DRQS 172614796,
    /// and the reversion of this workaround in DRQS 174079882.
    static void removePostAlwaysSignalsMitigation();
 
    /// Return `true` if  the mitigation of `post` always signalling the
    /// condition variable should be used, and `false` otherwise.
    static bool usePostAlwaysSignalsMitigation();
};
 
                       // ===========================
                       // class FastPostSemaphoreImpl
                       // ===========================
 
/// This class implements a semaphore type, optimized for `post`, for thread
/// synchronization.
///
/// See @ref bslmt_fastpostsemaphoreimpl 
template <class ATOMIC_OP, class MUTEX, class CONDITION, class THREADUTIL>
class FastPostSemaphoreImpl {
 
    // PRIVATE TYPES
    typedef typename ATOMIC_OP::AtomicTypes::Int64 AtomicInt64;
    typedef          bsls::Types::Int64            Int64;
 
    // DATA
    AtomicInt64 d_state;          // bit pattern representing the state of the
                                  // semaphore (see *Implementation* *Note*)
 
    MUTEX       d_waitMutex;      // mutex used with 'd_waitCondition', does
                                  // not protect any values
 
    CONDITION   d_waitCondition;  // condition variable for blocking/signalling
                                  // threads in the wait methods
 
    // PRIVATE CLASS METHODS
 
    /// Return a value suitable for detecting a rapid short sequence of
    /// `disable` and `enable` invocations, by comparing the value returned
    /// by `disabledGeneration` before and after the sequence, for the
    /// specified `state`.
    static bsls::Types::Int64 disabledGeneration(Int64 state);
 
    /// Return the semaphore value, without enforcing a zero minimum value,
    /// implied by the specified `state`.
    static bsls::Types::Int64 getValueRaw(Int64 state);
 
    /// Return `true` if the specified `state` implies the associated
    /// semaphore has available count.
    static bool hasAvailable(Int64 state);
 
    /// Return `true` if the specified `state` implies the associated
    /// semaphore has one or more threads blocked in a wait operation.
    static bool hasBlockedThread(Int64 state);
 
    /// Return `true` if the specified `state` implies the associated
    /// semaphore is "wait disabled".
    static bool isDisabled(Int64 state);
 
    /// Return `true` if the specified `state` implies the associated
    /// semaphore does not have sufficient resources to meet the demand upon
    /// the resources, implying there will be one or more threads blocked in
    /// wait operations (without further `post` invocations).
    static bool willHaveBlockedThread(Int64 state);
 
    // PRIVATE MANIPULATORS
 
    /// If this semaphore becomes disabled as detected from the disabled
    /// generation encoded in the specified `initialState` (see
    /// *Implementation* *Note*), return `e_DISABLED` with no effect on the
    /// count.  Otherwise, block until the count of this semaphore is a
    /// positive value or the specified `absTime` timeout expires.  If the
    /// count of this timed semaphore is a positive value, return 0 and
    /// atomically decrement the count.  If the `absTime` timeout expires,
    /// return `e_TIMED_OUT` with no effect on the count.  `absTime` is an
    /// absolute time represented as an interval from some epoch, which is
    /// determined by the clock indicated at construction (see {Supported
    /// Clock-Types} in the component-level documentation).  This method is
    /// invoked from `timedWait` when the invoking thread may have to be
    /// blocked.
    int timedWaitSlowPath(const bsls::TimeInterval& absTime,
                          const bsls::Types::Int64  initialState);
 
    /// If this semaphore becomes disabled as detected from the disabled
    /// generation encoded in the specified `initialState` (see
    /// *Implementation* *Note*), return `e_DISABLED` with no effect on the
    /// count.  Otherwise, block until the count of this semaphore is a
    /// positive value, return 0 and atomically decrement the count.  This
    /// method is invoked from `wait` when the invoking thread may have to
    /// be blocked.
    int waitSlowPath(const bsls::Types::Int64 initialState);
 
    // NOT IMPLEMENTED
    FastPostSemaphoreImpl(const FastPostSemaphoreImpl&);
    FastPostSemaphoreImpl& operator=(const FastPostSemaphoreImpl&);
 
  public:
    // PUBLIC CONSTANTS
    enum ReturnValue {
        e_SUCCESS     =  0,  // indicates success
        e_DISABLED    = -1,  // indicates semaphore is disabled
        e_TIMED_OUT   = -2,  // indicates operation timed out
        e_WOULD_BLOCK = -3,  // indicates operation would block ('tryWait')
        e_FAILED      = -4   // indicates failure reported from condition
    };
 
    // The following constants are used to maintain the semaphore`s `d_state`
    // attribute values for:
    // * number of threads blocked,
    // * generation count for tracking enabled/disabled,
    // * and available count.
    //
    // The `k_*_MASK` constants define the layout of the attributes, the
    // `k_*_INC` constants are used to modify the `d_state` attributes, and the
    // `k_*_SHIFT` constants allow recovery of the stored value.
    //
    // See **Implementation Note** for further details.  These are `public` to
    // ease testing.
 
    static const Int64 k_BLOCKED_INC        = 0x0000000000000001LL;
    static const Int64 k_BLOCKED_MASK       = 0x0000000000ffffffLL;
    static const Int64 k_DISABLED_GEN_INC   = 0x0000000001000000LL;
    static const Int64 k_DISABLED_GEN_MASK  = 0x000000000f000000LL;
    static const int   k_DISABLED_GEN_SHIFT = 24;
    static const Int64 k_AVAILABLE_INC      = 0x0000000010000000LL;
    static const Int64 k_AVAILABLE_MASK     = 0xfffffffff0000000LL;
    static const int   k_AVAILABLE_SHIFT    = 28;
 
    // CREATORS
 
    /// Create a `FastPostSemaphoreImpl` object initially having a count of 0.
    /// Optionally specify a `clockType` indicating the type of the system
    /// clock against which the `bsls::TimeInterval` `absTime` timeouts passed
    /// to the `timedWait` method are to be interpreted (see
    /// @ref bslmt_fastpostsemaphoreimpl-supported-clock-types .  If `clockType` is not specified then the
    /// realtime system clock is used.
    explicit
    FastPostSemaphoreImpl(
    bsls::SystemClockType::Enum clockType = bsls::SystemClockType::e_REALTIME);
 
    /// Create a `FastPostSemaphoreImpl` object initially having the
    /// specified `count`.  Optionally specify a `clockType` indicating the
    /// type of the system clock against which the `bsls::TimeInterval`
    /// `absTime` timeouts passed to the `timedWait` method are to be
    /// interpreted (see @ref bslmt_fastpostsemaphoreimpl-supported-clock-types .  If `clockType` is not
    /// specified then the realtime system clock is used.
    explicit
    FastPostSemaphoreImpl(
    int                         count,
    bsls::SystemClockType::Enum clockType = bsls::SystemClockType::e_REALTIME);
 
    /// Destroy this object.
     ~FastPostSemaphoreImpl() = default;
 
    // MANIPULATORS
 
    /// Disable waiting on this semaphore.  All subsequent invocations of
    /// `wait`, `tryWait`, and `timedWait` will fail immediately.  All
    /// blocked invocations of `wait` and `timedWait` will fail immediately.
    /// If the semaphore is initially disabled, this call has no effect.
    void disable();
 
    /// Enable waiting on this semaphore.  If the semaphore is initially
    /// enabled, this call has no effect.
    void enable();
 
    /// Atomically increment the count of this semaphore.
    void post();
 
    /// Atomically increase the count of this semaphore by the specified
    /// `value`.  The behavior is undefined unless `value > 0`.
    void post(int value);
 
    /// Atomically increase the count of this semaphore by the specified
    /// `value`.  If the resources available to this semaphore is greater
    /// than or equal to the specified `available` and the number of threads
    /// blocked in this semaphore is greater than or equal to the specified
    /// `blocked`, always send a signal to potentially wake a waiting thread
    /// (even if the signal should not be needed).  The behavior is
    /// undefined unless `value > 0`.  Note that this method is provided to
    /// help mitigate issues in the implementation of underlying
    /// synchronization primitives.
    void postWithRedundantSignal(int value, int available, int blocked);
 
    /// If the count of this semaphore is positive, reduce the count by the
    /// lesser of the count and the specified `maximumToTake` and return the
    /// magnitude of the change to the count.  Otherwise, do nothing and
    /// return 0.
    int take(int maximumToTake);
 
    /// If the count of this semaphore is positive, reduce the count to 0
    /// and return the original value of the count.  Otherwise, do nothing
    /// and return 0.
    int takeAll();
 
    /// If this semaphore is initially disabled, or becomes disabled while
    /// blocking, return `e_DISABLED` with no effect on the count.
    /// Otherwise, block until the count of this semaphore is a positive
    /// value or the specified `absTime` timeout expires.  If the count of
    /// this semaphore is a positive value, return 0 and atomically
    /// decrement the count.  If the `absTime` timeout expires, return
    /// `e_TIMED_OUT` with no effect on the count.  Return `e_FAILED` if an
    /// error occurs.  `absTime` is an *absolute* time represented as an
    /// interval from some epoch, which is determined by the clock indicated
    /// at construction (see {Supported Clock-Types} in the component
    /// documentation).
    int timedWait(const bsls::TimeInterval& absTime);
 
    /// If this semaphore is initially disabled, return `e_DISABLED` with no
    /// effect on the count.  Otherwise, if the count of this semaphore is a
    /// positive value, return 0 and atomically decrement the count.  If
    /// this semaphore is not disabled and the count of this semaphore is
    /// not a positive value, return `e_WOULD_BLOCK` with no effect on the
    /// count.
    int tryWait();
 
    /// If this semaphore is initially disabled, or becomes disabled while
    /// blocking, return `e_DISABLED` with no effect on the count.
    /// Otherwise, block until the count of this semaphore is a positive
    /// value, return 0 and atomically decrement the count.  Return
    /// `e_FAILED` if an error occurs.
    int wait();
 
    // ACCESSORS
 
    /// Return the clock type used for timeouts.
    bsls::SystemClockType::Enum clockType() const;
 
    /// Return an odd value if this semaphore is wait disabled, and an even
    /// value otherwise.  The returned value can be used to detect a rapid
    /// short sequence of `disable` and `enable` invocations by comparing
    /// the value returned by `getDisabledState` before and after the
    /// sequence.  For example, for any initial state of a semphore instance
    /// `obj`:
    /// @code
    /// int state = obj.getDisabledState();
    /// obj.disable();
    /// obj.enable();
    /// ASSERT(state != obj.getDisabledState());
    /// @endcode
    /// This functionality is useful in higher-level components to determine
    /// if this semaphore was disabled during an operation.
    int getDisabledState() const;
 
    /// Return the current value (`count > 0 ? count : 0`) of this
    /// semaphore.
    int getValue() const;
 
    /// Return `true` if this semaphore is wait disabled, and `false`
    /// otherwise.  Note that the semaphore is created in the "wait enabled"
    /// state.
    bool isDisabled() const;
};
 
// ============================================================================
//                             INLINE DEFINITIONS
// ============================================================================
 
                // -----------------------------------------
                // class FastPostSemaphoreImplWorkaroundUtil
                // -----------------------------------------
 
// PUBLIC CLASS METHODS
inline
void FastPostSemaphoreImplWorkaroundUtil::removePostAlwaysSignalsMitigation()
{
    s_postAlwaysSignals = false;
}
 
inline
bool FastPostSemaphoreImplWorkaroundUtil::usePostAlwaysSignalsMitigation()
{
    return s_postAlwaysSignals;
}
 
                       // ---------------------------
                       // class FastPostSemaphoreImpl
                       // ---------------------------
 
// PRIVATE CLASS METHODS
template <class ATOMIC_OP, class MUTEX, class CONDITION, class THREADUTIL>
inline
bsls::Types::Int64
                 FastPostSemaphoreImpl<ATOMIC_OP, MUTEX, CONDITION, THREADUTIL>
                                              ::disabledGeneration(Int64 state)
{
    return state & k_DISABLED_GEN_MASK;
}
 
template <class ATOMIC_OP, class MUTEX, class CONDITION, class THREADUTIL>
inline
bsls::Types::Int64
                 FastPostSemaphoreImpl<ATOMIC_OP, MUTEX, CONDITION, THREADUTIL>
                                                     ::getValueRaw(Int64 state)
{
    return (state >> k_AVAILABLE_SHIFT) - (state & k_BLOCKED_MASK);
}
 
template <class ATOMIC_OP, class MUTEX, class CONDITION, class THREADUTIL>
inline
bool FastPostSemaphoreImpl<ATOMIC_OP, MUTEX, CONDITION, THREADUTIL>
                                                    ::hasAvailable(Int64 state)
{
    return k_AVAILABLE_INC <= state;
}
 
template <class ATOMIC_OP, class MUTEX, class CONDITION, class THREADUTIL>
inline
bool FastPostSemaphoreImpl<ATOMIC_OP, MUTEX, CONDITION, THREADUTIL>
                                                ::hasBlockedThread(Int64 state)
{
    return 0 != (state & k_BLOCKED_MASK);
}
 
template <class ATOMIC_OP, class MUTEX, class CONDITION, class THREADUTIL>
inline
bool FastPostSemaphoreImpl<ATOMIC_OP, MUTEX, CONDITION, THREADUTIL>
                                                      ::isDisabled(Int64 state)
{
    return 0 != (state & k_DISABLED_GEN_INC);
}
 
template <class ATOMIC_OP, class MUTEX, class CONDITION, class THREADUTIL>
inline
bool FastPostSemaphoreImpl<ATOMIC_OP, MUTEX, CONDITION, THREADUTIL>
                                           ::willHaveBlockedThread(Int64 state)
{
    return (state >> k_AVAILABLE_SHIFT) < (state & k_BLOCKED_MASK);
}
 
// PRIVATE MANIPULATORS
template <class ATOMIC_OP, class MUTEX, class CONDITION, class THREADUTIL>
int FastPostSemaphoreImpl<ATOMIC_OP, MUTEX, CONDITION, THREADUTIL>
                    ::timedWaitSlowPath(const bsls::TimeInterval& absTime,
                                        const bsls::Types::Int64  initialState)
{
    int rv = e_SUCCESS;
 
    const Int64 disabledGen = disabledGeneration(initialState);
 
    // 'state' currently indicates the thread should block, yield and retest
    // instead
 
    THREADUTIL::yield();
 
    Int64 state = ATOMIC_OP::getInt64Acquire(&d_state);
 
    if (willHaveBlockedThread(state)) {
        {
            LockGuard<MUTEX> guard(&d_waitMutex);
 
            // note that the following operation indicates the thread is
            // blocked (see *Implementation* *Notes*) and does not affect the
            // decision for other threads to block
 
            state = ATOMIC_OP::addInt64NvAcqRel(
                                              &d_state,
                                              k_AVAILABLE_INC + k_BLOCKED_INC);
 
            // wait until there is an available resource or this semaphore is
            // disabled
 
            while (   !hasAvailable(state)
                   && disabledGen == disabledGeneration(state)) {
                int rv = d_waitCondition.timedWait(&d_waitMutex, absTime);
                if (rv) {
                    ATOMIC_OP::addInt64AcqRel(&d_state, -k_BLOCKED_INC);
                    if (-1 == rv) {
                        return e_TIMED_OUT;                           // RETURN
                    }
                    return e_FAILED;                                  // RETURN
                }
                state = ATOMIC_OP::getInt64Acquire(&d_state);
            }
 
            if (hasAvailable(state)) {
                state = ATOMIC_OP::addInt64NvAcqRel(
                                           &d_state,
                                           -(k_AVAILABLE_INC + k_BLOCKED_INC));
            }
            else {
                ATOMIC_OP::addInt64AcqRel(&d_state, -k_BLOCKED_INC);
                rv = e_DISABLED;
            }
        }
 
        // signal when 'state' indicates there is an available resource, the
        // semaphore is not disabled, and there are blocked threads
 
        if (   hasAvailable(state)
            && !isDisabled(state)
            && hasBlockedThread(state)) {
            d_waitCondition.signal();
        }
    }
    else {
        // signal when 'state' indicates there is an available resource, the
        // semaphore is not disabled, and there are blocked threads
 
        if (   hasAvailable(state)
            && !isDisabled(state)
            && hasBlockedThread(state)) {
            {
                LockGuard<MUTEX> guard(&d_waitMutex);
            }
            d_waitCondition.signal();
        }
    }
 
    return rv;
}
 
template <class ATOMIC_OP, class MUTEX, class CONDITION, class THREADUTIL>
int FastPostSemaphoreImpl<ATOMIC_OP, MUTEX, CONDITION, THREADUTIL>
                          ::waitSlowPath(const bsls::Types::Int64 initialState)
{
    int rv = e_SUCCESS;
 
    const Int64 disabledGen = disabledGeneration(initialState);
 
    // 'state' currently indicates the thread should block, yield and retest
    // instead
 
    THREADUTIL::yield();
 
    Int64 state = ATOMIC_OP::getInt64Acquire(&d_state);
 
    if (willHaveBlockedThread(state)) {
        {
            LockGuard<MUTEX> guard(&d_waitMutex);
 
            // note that the following operation indicates the thread is
            // blocked (see *Implementation* *Notes*) and does not affect the
            // decision for other threads to block
 
            state = ATOMIC_OP::addInt64NvAcqRel(
                                              &d_state,
                                              k_AVAILABLE_INC + k_BLOCKED_INC);
 
            // wait until there is an available resource or this semaphore is
            // disabled
 
            while (   !hasAvailable(state)
                   && disabledGen == disabledGeneration(state)) {
                int rv = d_waitCondition.wait(&d_waitMutex);
                if (rv) {
                    ATOMIC_OP::addInt64AcqRel(&d_state, -k_BLOCKED_INC);
                    return e_FAILED;                                  // RETURN
                }
                state = ATOMIC_OP::getInt64Acquire(&d_state);
            }
 
            if (hasAvailable(state)) {
                state = ATOMIC_OP::addInt64NvAcqRel(
                                           &d_state,
                                           -(k_AVAILABLE_INC + k_BLOCKED_INC));
            }
            else {
                ATOMIC_OP::addInt64AcqRel(&d_state, -k_BLOCKED_INC);
                rv = e_DISABLED;
            }
        }
 
        // signal when 'state' indicates there is an available resource, the
        // semaphore is not disabled, and there are blocked threads
 
        if (   hasAvailable(state)
            && !isDisabled(state)
            && hasBlockedThread(state)) {
            d_waitCondition.signal();
        }
    }
    else {
        // signal when 'state' indicates there is an available resource, the
        // semaphore is not disabled, and there are blocked threads
 
        if (   hasAvailable(state)
            && !isDisabled(state)
            && hasBlockedThread(state)) {
            {
                LockGuard<MUTEX> guard(&d_waitMutex);
            }
            d_waitCondition.signal();
        }
    }
 
    return rv;
}
 
// CREATORS
template <class ATOMIC_OP, class MUTEX, class CONDITION, class THREADUTIL>
inline
FastPostSemaphoreImpl<ATOMIC_OP, MUTEX, CONDITION, THREADUTIL>
                 ::FastPostSemaphoreImpl(bsls::SystemClockType::Enum clockType)
: d_waitMutex()
, d_waitCondition(clockType)
{
    ATOMIC_OP::initInt64(&d_state, 0);
}
 
template <class ATOMIC_OP, class MUTEX, class CONDITION, class THREADUTIL>
inline
FastPostSemaphoreImpl<ATOMIC_OP, MUTEX, CONDITION, THREADUTIL>
                 ::FastPostSemaphoreImpl(int                         count,
                                         bsls::SystemClockType::Enum clockType)
: d_waitMutex()
, d_waitCondition(clockType)
{
    ATOMIC_OP::initInt64(&d_state, k_AVAILABLE_INC * count);
}
 
// MANIPULATORS
template <class ATOMIC_OP, class MUTEX, class CONDITION, class THREADUTIL>
void FastPostSemaphoreImpl<ATOMIC_OP, MUTEX, CONDITION, THREADUTIL>::disable()
{
    Int64 state = ATOMIC_OP::getInt64Acquire(&d_state);
 
    while (false == isDisabled(state)) {
        const Int64 expState = state;
 
        // increment, without overflowing, the disabled attribute
 
        Int64 newState = (state & ~k_DISABLED_GEN_MASK) |
                          ((state + k_DISABLED_GEN_INC) & k_DISABLED_GEN_MASK);
 
        state = ATOMIC_OP::testAndSwapInt64AcqRel(&d_state,
                                                  state,
                                                  newState);
 
        if (expState == state) {
            state = newState;
 
            // note that 'd_waitMutex' must be acquired to ensure a thread in a
            // wait operation either "sees" the change in state before
            // determining whether to block using 'd_waitCondition', or has
            // blocked and will receive a signal sent to 'd_waitCondition'
 
            {
                LockGuard<MUTEX> guard(&d_waitMutex);
            }
            d_waitCondition.broadcast();
        }
    }
 
    // When threads blocked on 'd_waitCondition' are signalled, they must
    // prefer consuming available resources and returning success over
    // returning 'e_DISABLED'.  This ensures the signalled thread obtains the
    // resource when another thread issues a post followed by a disablement
    // (see DRQS 153332608).  Similarly, to ensure a disablement followed by a
    // post method does not provide a resource to the signalled thread,
    // 'disable' must wait until no threads will block before returning.  Note
    // that the semaphore may be re-enabled (and re-disabled) during this wait.
 
    while (isDisabled(state) && willHaveBlockedThread(state)) {
        THREADUTIL::yield();
 
        state = ATOMIC_OP::getInt64Acquire(&d_state);
    }
}
 
template <class ATOMIC_OP, class MUTEX, class CONDITION, class THREADUTIL>
void FastPostSemaphoreImpl<ATOMIC_OP, MUTEX, CONDITION, THREADUTIL>::enable()
{
    Int64 state = ATOMIC_OP::getInt64Acquire(&d_state);
 
    while (isDisabled(state)) {
        // When this semaphore is disabled, an 'enable' followed by a post
        // method must not provide a resource to a signalled thread that should
        // return 'e_DISABLED' (see note in 'disable').  Hence, wait until no
        // threads will block before performing the enablement.
 
        if (willHaveBlockedThread(state)) {
            THREADUTIL::yield();
 
            state = ATOMIC_OP::getInt64Acquire(&d_state);
        }
        else {
            const Int64 expState = state;
 
            // increment, without overflowing, the disabled attribute
 
            Int64 newState = (state & ~k_DISABLED_GEN_MASK) |
                          ((state + k_DISABLED_GEN_INC) & k_DISABLED_GEN_MASK);
 
            state = ATOMIC_OP::testAndSwapInt64AcqRel(&d_state,
                                                      state,
                                                      newState);
 
            if (expState == state) {
                state = newState;
            }
        }
    }
}
 
template <class ATOMIC_OP, class MUTEX, class CONDITION, class THREADUTIL>
inline
void FastPostSemaphoreImpl<ATOMIC_OP, MUTEX, CONDITION, THREADUTIL>::post()
{
    Int64 state = ATOMIC_OP::addInt64NvAcqRel(&d_state, k_AVAILABLE_INC);
 
    // signal only when 'state' indicates there are no other threads that can
    // unblock blocked threads
 
    if (   (   FastPostSemaphoreImplWorkaroundUtil::
                                               usePostAlwaysSignalsMitigation()
            || k_AVAILABLE_INC == (state & k_AVAILABLE_MASK))
        && !isDisabled(state)
        && hasBlockedThread(state)) {
 
        // note that 'd_waitMutex' must be acquired to ensure a thread in a
        // wait operation either "sees" the change in state before determining
        // whether to block using 'd_waitCondition', or has blocked and will
        // receive a signal sent to 'd_waitCondition'
 
        {
            LockGuard<MUTEX> guard(&d_waitMutex);
        }
        d_waitCondition.signal();
    }
}
 
template <class ATOMIC_OP, class MUTEX, class CONDITION, class THREADUTIL>
inline
void FastPostSemaphoreImpl<ATOMIC_OP, MUTEX, CONDITION, THREADUTIL>
                                                              ::post(int value)
{
    Int64 v     = k_AVAILABLE_INC * value;
    Int64 state = ATOMIC_OP::addInt64NvAcqRel(&d_state, v);
 
    // signal only when 'state' indicates there are no other threads that can
    // unblock blocked threads
 
    if (   (   FastPostSemaphoreImplWorkaroundUtil::
                                               usePostAlwaysSignalsMitigation()
            || v == (state & k_AVAILABLE_MASK))
        && !isDisabled(state)
        && hasBlockedThread(state)) {
 
        // note that 'd_waitMutex' must be acquired to ensure a thread in a
        // wait operation either "sees" the change in state before determining
        // whether to block using 'd_waitCondition', or has blocked and will
        // receive a signal sent to 'd_waitCondition'
 
        {
            LockGuard<MUTEX> guard(&d_waitMutex);
        }
        d_waitCondition.signal();
    }
}
 
template <class ATOMIC_OP, class MUTEX, class CONDITION, class THREADUTIL>
inline
void FastPostSemaphoreImpl<ATOMIC_OP, MUTEX, CONDITION, THREADUTIL>
                                       ::postWithRedundantSignal(int value,
                                                                 int available,
                                                                 int blocked)
{
    Int64 v     = k_AVAILABLE_INC * value;
    Int64 state = ATOMIC_OP::addInt64NvAcqRel(&d_state, v);
 
    // signal only when 'state' indicates there are no other threads that can
    // unblock blocked threads, or there are 'available' or more resources and
    // 'blocked' or more threads
 
    if (   (   FastPostSemaphoreImplWorkaroundUtil::
                                               usePostAlwaysSignalsMitigation()
            || v == (state & k_AVAILABLE_MASK)
            || (   k_AVAILABLE_INC * available <= (state & k_AVAILABLE_MASK)
                && blocked <= (state & k_BLOCKED_MASK)))
        && !isDisabled(state)
        && hasBlockedThread(state)) {
 
        // note that 'd_waitMutex' must be acquired to ensure a thread in a
        // wait operation either "sees" the change in state before determining
        // whether to block using 'd_waitCondition', or has blocked and will
        // receive a signal sent to 'd_waitCondition'
 
        {
            LockGuard<MUTEX> guard(&d_waitMutex);
        }
        d_waitCondition.signal();
 
        BSLS_REVIEW_OPT(   (   FastPostSemaphoreImplWorkaroundUtil::
                                               usePostAlwaysSignalsMitigation()
                            || v == (state & k_AVAILABLE_MASK))
                        && "redundant signal sent");
    }
}
 
template <class ATOMIC_OP, class MUTEX, class CONDITION, class THREADUTIL>
inline
int FastPostSemaphoreImpl<ATOMIC_OP, MUTEX, CONDITION, THREADUTIL>
                                                      ::take(int maximumToTake)
{
    Int64 state = ATOMIC_OP::getInt64Acquire(&d_state);
    Int64 expState;
    Int64 count;
 
    do {
        // remove all available up to 'maximumToTake'
 
        expState = state;
 
        count = getValueRaw(state);
 
        if (0 >= count) {
            return 0;                                                 // RETURN
        }
 
        if (maximumToTake < count) {
            count = maximumToTake;
        }
 
        state = ATOMIC_OP::testAndSwapInt64AcqRel(
                                              &d_state,
                                              state,
                                              state - k_AVAILABLE_INC * count);
    } while (state != expState);
 
    return static_cast<int>(count);
}
 
template <class ATOMIC_OP, class MUTEX, class CONDITION, class THREADUTIL>
inline
int FastPostSemaphoreImpl<ATOMIC_OP, MUTEX, CONDITION, THREADUTIL>::takeAll()
{
    return take(INT_MAX);
}
 
template <class ATOMIC_OP, class MUTEX, class CONDITION, class THREADUTIL>
inline
int FastPostSemaphoreImpl<ATOMIC_OP, MUTEX, CONDITION, THREADUTIL>
                                 ::timedWait(const bsls::TimeInterval& absTime)
{
    Int64 state = ATOMIC_OP::addInt64NvAcqRel(&d_state, -k_AVAILABLE_INC);
 
    if (isDisabled(state)) {
        post();
        return e_DISABLED;                                            // RETURN
    }
 
    if (willHaveBlockedThread(state)) {
        return timedWaitSlowPath(absTime, state);                     // RETURN
    }
 
    return 0;
}
 
template <class ATOMIC_OP, class MUTEX, class CONDITION, class THREADUTIL>
inline
int FastPostSemaphoreImpl<ATOMIC_OP, MUTEX, CONDITION, THREADUTIL>::tryWait()
{
    Int64 state = ATOMIC_OP::addInt64NvAcqRel(&d_state, -k_AVAILABLE_INC);
 
    if (isDisabled(state)) {
        post();
        return e_DISABLED;                                            // RETURN
    }
 
    if (willHaveBlockedThread(state)) {
        post();
        return e_WOULD_BLOCK;                                         // RETURN
    }
 
    return 0;
}
 
template <class ATOMIC_OP, class MUTEX, class CONDITION, class THREADUTIL>
inline
int FastPostSemaphoreImpl<ATOMIC_OP, MUTEX, CONDITION, THREADUTIL>::wait()
{
    Int64 state = ATOMIC_OP::addInt64NvAcqRel(&d_state, -k_AVAILABLE_INC);
 
    if (isDisabled(state)) {
        post();
        return e_DISABLED;                                            // RETURN
    }
 
    if (willHaveBlockedThread(state)) {
        return waitSlowPath(state);                                   // RETURN
    }
 
    return 0;
}
 
// ACCESSORS
template <class ATOMIC_OP, class MUTEX, class CONDITION, class THREADUTIL>
inline
bsls::SystemClockType::Enum
FastPostSemaphoreImpl<ATOMIC_OP, MUTEX, CONDITION, THREADUTIL>
                                                            ::clockType() const
{
    return d_waitCondition.clockType();
}
 
template <class ATOMIC_OP, class MUTEX, class CONDITION, class THREADUTIL>
inline
int FastPostSemaphoreImpl<ATOMIC_OP, MUTEX, CONDITION, THREADUTIL>
                                                     ::getDisabledState() const
{
    Int64 state = ATOMIC_OP::getInt64Acquire(&d_state);
 
    return static_cast<int>((state & k_DISABLED_GEN_MASK)
                                                      >> k_DISABLED_GEN_SHIFT);
}
 
template <class ATOMIC_OP, class MUTEX, class CONDITION, class THREADUTIL>
inline
int FastPostSemaphoreImpl<ATOMIC_OP, MUTEX, CONDITION, THREADUTIL>
                                                             ::getValue() const
{
    Int64 count = getValueRaw(ATOMIC_OP::getInt64Acquire(&d_state));
 
    return static_cast<int>(count > 0 ? count : 0);
}
 
template <class ATOMIC_OP, class MUTEX, class CONDITION, class THREADUTIL>
inline
bool FastPostSemaphoreImpl<ATOMIC_OP, MUTEX, CONDITION, THREADUTIL>
                                                           ::isDisabled() const
{
    Int64 state = ATOMIC_OP::getInt64Acquire(&d_state);
 
    return isDisabled(state);
}
 
}  // close package namespace
 
 
#endif
 
// ----------------------------------------------------------------------------
// Copyright 2020 Bloomberg Finance L.P.
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
//     http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.
// ----------------------------- END-OF-FILE ----------------------------------
 
/** @} */
/** @} */
/** @} */