Merge Atomics branch into Master

7 views
Skip to first unread message

Jeffrey Walton

unread,
Jun 6, 2016, 7:35:07 AM6/6/16
to Crypto++ Users
Hi Everyone,

Crypto++ provides a Singleton class in misc.h. The original class suffered C++03 (and earlier) shortcomings. A design compromise was made to leak memory rather than using locks to increase portability.

We now have a portable implementation that does not depend on non-portable locks. It follows http://preshing.com/20130930/double-checked-locking-is-fixed-in-cpp11.

template <class T, class F, int instance>
  const T & Singleton<T, F, instance>::Ref(CRYPTOPP_NOINLINE_DOTDOTDOT) const
{
    static std::mutex s_mutex;
    static std::atomic<T*> s_pObject;

    T *p = s_pObject.load(std::memory_order_relaxed);
    std::atomic_thread_fence(std::memory_order_acquire);

    if (p)
        return *p;

    std::lock_guard<std::mutex> lock(s_mutex);
    p = s_pObject.load(std::memory_order_relaxed);
    std::atomic_thread_fence(std::memory_order_acquire);

    if (p)
        return *p;

    T *newObject = m_objectFactory();
    s_pObject.store(newObject, std::memory_order_relaxed);
    std::atomic_thread_fence(std::memory_order_release);

    return *newObject;
}

The implementation is guarded by both CRYPTOPP_CXX11_ATOMICS and CRYPTOPP_CXX11_SYNCHRONIZATION because atomics is not a proper subset of synchronization when it comes to compiler support and versions.

I'd like to merge it once it goes through a round of testing.

Are there any objections?

Jeff

Jeffrey Walton

unread,
Jun 6, 2016, 10:09:53 PM6/6/16
to Crypto++ Users
Reply all
Reply to author
Forward
0 new messages