Performance defect in disruptorplus::spin_wait ctor

[pkarneliuk]

The constructor of class disruptorplus::spin_wait has the call to std::thread::hardware_concurrency(). It affects latency of all
disruptorplus::spin_wait_strategy::wait_until_published() methods:

disruptorplus/include/disruptorplus/spin_wait.hpp

Lines 40 to 50 in 748e354

	spin_wait()
	{
	reset();
	}
	/// \brief
	/// Reset the spin_wait back to its original state.
	void reset()
	{
	m_value = std::thread::hardware_concurrency() > 1 ? 0 : 10;
	}

disruptorplus/include/disruptorplus/spin_wait_strategy.hpp

Lines 51 to 65 in 748e354

	sequence_t wait_until_published(
	sequence_t sequence,
	size_t count,
	const std::atomic<sequence_t>* const sequences[])
	{
	assert(count > 0);
	spin_wait spinner;
	sequence_t result = minimum_sequence_after(sequence, count, sequences);
	while (difference(result, sequence) < 0)
	{
	spinner.spin_once();
	result = minimum_sequence_after(sequence, count, sequences);
	}
	return result;
	}

disruptorplus/include/disruptorplus/spin_wait_strategy.hpp

Lines 142 to 163 in 748e354

	template<typename Clock, typename Duration>
	sequence_t wait_until_published(
	sequence_t sequence,
	size_t count,
	const std::atomic<sequence_t>* const sequences[],
	const std::chrono::time_point<Clock, Duration>& timeoutTime)
	{
	assert(count > 0);
	spin_wait spinner;
	sequence_t result = minimum_sequence_after(sequence, count, sequences);
	while (difference(result, sequence) < 0)
	{
	if (spinner.next_spin_will_yield() && timeoutTime < Clock::now())
	{
	// Out of time.
	return result;
	}
	spinner.spin_once();
	result = minimum_sequence_after(sequence, count, sequences);
	}
	return result;
	}

Also, there are some redundant code which trying to hide negative impact of this defect:

disruptorplus/include/disruptorplus/sequence_barrier_group.hpp

Lines 164 to 170 in 748e354

	size_t count = m_sequences.size();
	sequence_t current = minimum_sequence_after(sequence, count, m_sequences.data());
	if (difference(current, sequence) >= 0)
	{
	return current;
	}

disruptorplus/include/disruptorplus/sequence_barrier_group.hpp

Lines 204 to 210 in 748e354

	size_t count = m_sequences.size();
	sequence_t current = minimum_sequence_after(sequence, count, m_sequences.data());
	if (difference(current, sequence) >= 0)
	{
	return current;
	}

Possible fix is to save the result of std::thread::hardware_concurrency() in some static variable and reuse it in spin_wait ctor.