bool
test_bitops(boost::atomic<value_type> & shared_value, size_t instance)
{
    size_t shift = instance * 8;
    value_type mask = 0xff << shift;

    value_type expected = 0;

    for (size_t k = 0; k < 8; k++) {
        value_type mod = 1 << k;
        value_type tmp = shared_value.fetch_or(mod << shift, boost::memory_order_relaxed);
        if ( (tmp & mask) != (expected << shift))
            return false;
        expected = expected | mod;
    }
    for (size_t k = 0; k < 8; k++) {
        value_type tmp = shared_value.fetch_and( ~ (1 << (shift + k)), boost::memory_order_relaxed);
        if ( (tmp & mask) != (expected << shift))
            return false;
        expected = expected & ~(1<<k);
    }
    for (size_t k = 0; k < 8; k++) {
        value_type mod = 255 ^ (1 << k);
        value_type tmp = shared_value.fetch_xor(mod << shift, boost::memory_order_relaxed);
        if ( (tmp & mask) != (expected << shift))
            return false;
        expected = expected ^ mod;
    }

    value_type tmp = shared_value.fetch_and( ~mask, boost::memory_order_relaxed);
    if ( (tmp & mask) != (expected << shift) )
        return false;

    return true;
}

namespace Common {

UString generateIDRandomString() {
	std::stringstream ss;

	ss << boost::uuids::random_generator()();

	return ss.str();
}

static boost::atomic<uint32> idNumber(1);
uint32 generateIDNumber() {
	return idNumber.fetch_add(1);
}

static UString uint64ToString(uint64 i) {
	std::string str;
	str.reserve(20);

	do {
		str += "0123456789"[i % 10];
		i /= 10;
	} while (i);

	std::reverse(str.begin(), str.end());

	return str;
}

static boost::atomic<uint64> idNumberString(1);
UString generateIDNumberString() {
	return uint64ToString(idNumberString.fetch_add(1));
}

} // End of namespace Common

 inline T get_and_reset_value(boost::atomic<T>& value, bool reset)
 {
     T result = value.load();
     if (reset)
         value.store(0);
     return result;
 }

/**
 * The environment should be initialised exactly once during shared library initialisation.
 * Prior to that, all operations work in pass-through mode.
 * To overcome undefined global initialisation order, we use a local static variable.
 * This variable is accesd by a single writer (lib (de)initialisation), and multiple readers (exported operations).
 * Writer: toggle == true (toggles is_initialised)
 * Reader: toggle == false (returns the current value of is_initialised)
 *
 * NOTE: pthread_once initialisation for the overlay leads to a deadlock due to a re-entry situation:
 * - environment ctor is called the first time via pthread_once
 * - which leads a write somewhere (probably to a socket during client initialisation)
 * - which then enters the pthread_once initialisation mechanism again and blocks deadlocks
 */
bool overlay_initialized(bool toggle /* defaults to: false */) {
  static boost::atomic<bool> is_initilized(false);

  // writer
  if(toggle) {
    return is_initilized.exchange(!is_initilized.load());
  }

  // reader
  return is_initilized.load();
}

void plain_arguments(Executor& exec)
{
    void_f4_count.store(0);
    int_f4_count.store(0);

    {
        future<void> f1 = dataflow(exec, &void_f4, 42);
        future<int> f2 = dataflow(exec, &int_f4, 42);

        f1.wait();
        HPX_TEST_EQ(void_f4_count, 1u);

        HPX_TEST_EQ(f2.get(), 84);
        HPX_TEST_EQ(int_f4_count, 1u);
    }

    void_f5_count.store(0);
    int_f5_count.store(0);

    {
        future<void> f1 = dataflow(exec, &void_f5, 42, async(&int_f));
        future<int> f2 = dataflow(exec, &int_f5, 42, async(&int_f));

        f1.wait();
        HPX_TEST_EQ(void_f5_count, 1u);

        HPX_TEST_EQ(f2.get(), 126);
        HPX_TEST_EQ(int_f5_count, 1u);
    }
}

void plain_deferred_arguments()
{
    void_f4_count.store(0);
    int_f4_count.store(0);

    {
        future<void> f1 = dataflow(hpx::launch::deferred, &void_f4, 42);
        future<int> f2 = dataflow(hpx::launch::deferred, &int_f4, 42);

        f1.wait();
        HPX_TEST_EQ(void_f4_count, 1u);

        HPX_TEST_EQ(f2.get(), 84);
        HPX_TEST_EQ(int_f4_count, 1u);
    }

    void_f5_count.store(0);
    int_f5_count.store(0);

    {
        future<void> f1 = dataflow(&void_f5, 42, async(hpx::launch::deferred, &int_f));
        future<int> f2 = dataflow(&int_f5, 42, async(hpx::launch::deferred, &int_f));

        f1.wait();
        HPX_TEST_EQ(void_f5_count, 1u);

        HPX_TEST_EQ(f2.get(), 126);
        HPX_TEST_EQ(int_f5_count, 1u);
    }
}

void	LoggingStreamImpl::UpdateIsLogging() const
{
    m_lastLoggingFilterChangeCount.store( g_LoggingFilterChangeCount.load() );

    boost::log::record		testRecord = m_logger.open_record( boost::log::keywords::severity = m_severityLevel );

    m_isLogging.store( (bool)testRecord );
}

// run an inlet for some time (optionally with sporadic interruptions in between)
void run_inlet(const double duration_=0.0, const string name_=string(), const string type_=string(), const int in_chunks_=-1, const int request_info_=-1, const int request_time_=-1, const double seconds_between_failures_=0.0) {
	num_inlets.fetch_add(1);
	std::ostringstream s; s << boost::this_thread::get_id();
	srand((unsigned)boost::hash<string>()(s.str()));
	try {
		// choose random parameters if desired		
		double duration = (duration_ == 0.0) ? 1.0+rand()%(max_outlet_duration-1) : duration_;
		string name = name_.empty() ? names[rand()%(sizeof(names)/sizeof(names[0]))] : name_;
		string type = type_.empty() ? types[rand()%(sizeof(types)/sizeof(types[0]))] : type_;
		int request_info = (request_info_==-1) ? rand()%3 == 0 : request_info_;
		int request_time = (request_time_==-1) ? rand()%3 == 0 : request_time_;
		double seconds_between_failures = (seconds_between_failures_ == 0.0) ? (inlet_min_failure_interval_ms+rand()%outlet_max_failure_interval_ms)/1000.0 : seconds_between_failures_;

		// resolve by type...
		vector<lsl::stream_info> results = lsl::resolve_stream("type",type,1,5);
		if (results.empty())
			throw lsl::lost_error("No stream found.");
		lsl::stream_info result = results[rand()%results.size()];
		vector<float> chunk;		

		// and run...
		double t=0.0;
		for (double endtime = lsl::local_clock()+duration;lsl::local_clock()<endtime;) {
			// run a single execution of the inlet
			{
				cout << "new inlet(" << name << "," << type << ")...";
				lsl::stream_inlet inlet(result,max_buffered);
				cout << "done." << endl;
				int numchans = inlet.info().channel_count();
				init_sample(numchans*(int)ceil(max_chunk_len_ms*result.nominal_srate()/1000*max_chunk_oversize_factor),chunk);
				if (request_info) {
					cout << "  info = " << inlet.info(1.0).name() << endl;
				}
				for (double fail_at = lsl::local_clock()+seconds_between_failures;lsl::local_clock()<fail_at;) {
					boost::this_thread::sleep(boost::posix_time::milliseconds(1+rand()%max_inlet_poll_interval_ms));
					inlet.pull_chunk_multiplexed(&chunk[0],NULL,chunk.size(),0);
					if (request_time)
						t = inlet.time_correction(1.0);
				}
			}
			cout << "del inlet(" << name << "," << type << ")" << endl;
			if (request_time)
				cout << "  tcorr = " << t << endl;
			// downtime
			boost::this_thread::sleep(boost::posix_time::millisec(100*(rand()%50)));
		}
	} 
	catch(lsl::timeout_error &) { std::cerr << "Timeout exceeded; stopping inlet." << std::endl; }
	catch(lsl::lost_error &) { std::cerr << "Found no matching outlet; stopping inlet." << std::endl; }
	catch(std::exception &e) {
		std::cerr << "ERROR during run_inlet() Stress-test function: " <<	e.what() << std::endl;
	}
	num_inlets.fetch_sub(1);
}

void future_function_pointers()
{
    future_void_f1_count.store(0);
    future_void_f2_count.store(0);

    future<void> f1
        = dataflow(
            &future_void_f1, async(&future_void_sf1,
                shared_future<void>(make_ready_future()))
        );

    f1.wait();

    HPX_TEST_EQ(future_void_f1_count, 2u);
    future_void_f1_count.store(0);

    future<void> f2 = dataflow(
        &future_void_f2
      , async(&future_void_sf1, shared_future<void>(make_ready_future()))
      , async(&future_void_sf1, shared_future<void>(make_ready_future()))
    );

    f2.wait();
    HPX_TEST_EQ(future_void_f1_count, 2u);
    HPX_TEST_EQ(future_void_f2_count, 1u);
    future_void_f1_count.store(0);
    future_void_f2_count.store(0);

    future<int> f3 = dataflow(
        &future_int_f1
      , make_ready_future()
    );

    HPX_TEST_EQ(f3.get(), 1);
    HPX_TEST_EQ(future_int_f1_count, 1u);
    future_int_f1_count.store(0);

    future<int> f4 = dataflow(
        &future_int_f2
      , dataflow(&future_int_f1, make_ready_future())
      , dataflow(&future_int_f1, make_ready_future())
    );

    HPX_TEST_EQ(f4.get(), 2);
    HPX_TEST_EQ(future_int_f1_count, 2u);
    HPX_TEST_EQ(future_int_f2_count, 1u);
    future_int_f1_count.store(0);
    future_int_f2_count.store(0);

    future_int_f_vector_count.store(0);
    std::vector<future<int> > vf;
    for(std::size_t i = 0; i < 10; ++i)
    {
        vf.push_back(dataflow(&future_int_f1, make_ready_future()));
    }
    future<int> f5 = dataflow(&future_int_f_vector, boost::ref(vf));

    HPX_TEST_EQ(f5.get(), 10);
}

    void call_void()
    {
        ++count_call_void;

        // make sure this function is not concurrently invoked
        HPX_TEST_EQ(count_active_call_void.fetch_add(1) + 1, 1);

        hpx::this_thread::suspend(boost::chrono::microseconds(100));

        --count_active_call_void;
        HPX_TEST_EQ(count_active_call_void.load(), 0);
    }

void init_globals()
{
    // Retrieve command line using the Boost.ProgramOptions library.
    boost::program_options::variables_map vm;
    if (!hpx::util::retrieve_commandline_arguments(get_commandline_options(), vm))
    {
        HPX_THROW_EXCEPTION(hpx::commandline_option_error,
            "fibonacci_futures_distributed",
            "failed to handle command line options");
        return;
    }

    boost::uint64_t n = vm["n-value"].as<boost::uint64_t>();

    threshold = vm["threshold"].as<unsigned int>();
    if (threshold < 2 || threshold > n) {
        HPX_THROW_EXCEPTION(hpx::commandline_option_error,
            "fibonacci_futures_distributed",
            "wrong command line argument value for option 'threshold', "
            "should be in between 2 and n-value, value specified: " +
                std::to_string(threshold));
        return;
    }

    distribute_at = vm["distribute-at"].as<unsigned int>();
    if (distribute_at < 2 || distribute_at > n) {
        HPX_THROW_EXCEPTION(hpx::commandline_option_error,
            "fibonacci_futures_distributed",
            "wrong command line argument value for option 'distribute-at', "
            "should be in between 2 and n-value, value specified: " +
                std::to_string(distribute_at));
        return;
    }

    here = hpx::find_here();
    next_locality.store(0);
    serial_execution_count.store(0);

    // try to more evenly distribute the work over the participating localities
    std::vector<hpx::id_type> locs = hpx::find_all_localities();
    std::size_t num_repeats = vm["loc-repeat"].as<int>();

    localities.push_back(here);      // add ourselves
    for (std::size_t j = 0; j != num_repeats; ++j)
    {
        for (std::size_t i = 0; i != locs.size(); ++i)
        {
            if (here == locs[i])
                continue;
            localities.push_back(locs[i]);
        }
    }
}

 bool try_basic_lock(thread_id_type current_thread_id)
 {
     if (mtx.try_lock())
     {
         locking_thread_id.exchange(current_thread_id);
         util::ignore_lock(&mtx);
         util::register_lock(this);
         recursion_count.store(1);
         return true;
     }
     return false;
 }

TORRENT_NO_RETURN TORRENT_EXPORT void assert_fail(char const* expr, int line
	, char const* file, char const* function, char const* value, int kind)
{
#ifdef TORRENT_PRODUCTION_ASSERTS
	// no need to flood the assert log with infinite number of asserts
	if (assert_counter.fetch_add(1) + 1 > 500) return;
#endif

	char stack[8192];
	stack[0] = '\0';
	print_backtrace(stack, sizeof(stack), 0);

	char const* message = "assertion failed. Please file a bugreport at "
		"http://code.google.com/p/libtorrent/issues\n"
		"Please include the following information:\n\n"
		"version: " LIBTORRENT_VERSION "\n"
		LIBTORRENT_REVISION "\n";

	switch (kind)
	{
		case 1:
			message = "A precondition of a libtorrent function has been violated.\n"
				"This indicates a bug in the client application using libtorrent\n";
	}
	  
	assert_print("%s\n"
#ifdef TORRENT_PRODUCTION_ASSERTS
		"#: %d\n"
#endif
		"file: '%s'\n"
		"line: %d\n"
		"function: %s\n"
		"expression: %s\n"
		"%s%s\n"
		"stack:\n"
		"%s\n"
		, message
#ifdef TORRENT_PRODUCTION_ASSERTS
		, assert_counter.load()
#endif
		, file, line, function, expr
		, value ? value : "", value ? "\n" : ""
		, stack);

	// if production asserts are defined, don't abort, just print the error
#ifndef TORRENT_PRODUCTION_ASSERTS
 	// send SIGINT to the current process
 	// to break into the debugger
 	raise(SIGINT);
 	abort();
#endif
}

// run an outlet for some time (optionally with sporadic interruptions in between)
void run_outlet(const double duration_=0.0, const string name_=string(), const string type_=string(), const int numchan_=0, const lsl::channel_format_t fmt_=lsl::cf_undefined, const double srate_=0.0, const double seconds_between_failures_=0.0, const int chunk_len_=0) {
	num_outlets.fetch_add(1);
	std::ostringstream s; s << boost::this_thread::get_id();
	srand((unsigned)boost::hash<string>()(s.str()));
	try {
		// choose random parameters if desired
		double duration = (duration_ == 0.0) ? 1.0+rand()%(max_outlet_duration-1) : duration_;
		string name = name_.empty() ? names[rand()%(sizeof(names)/sizeof(names[0]))] : name_;
		string type = type_.empty() ? types[rand()%(sizeof(types)/sizeof(types[0]))] : type_;
		int numchan = (numchan_ == 0) ? 1+(rand()%(max_channels-1)) : numchan_;
		double srate = (srate_ == 0.0) ? 1.0 + (rand()%(max_srate-1)) : srate_;
		lsl::channel_format_t fmt = (fmt_ == lsl::cf_undefined) ? fmts[rand()%6] : fmt_;
		double seconds_between_failures = (seconds_between_failures_ == 0.0) ? (inlet_min_failure_interval_ms+rand()%outlet_max_failure_interval_ms)/1000.0 : seconds_between_failures_;
		int chunk_len = (chunk_len_ == 0) ? std::max(min_chunk_len_ms,(rand()%max_chunk_len_ms)) : chunk_len_;

		// create a new streaminfo
		lsl::stream_info info(name,type,numchan,srate,fmt,boost::uuids::to_string(boost::uuids::random_generator()()));

		// initialize data to send
		vector<float> chunk;
		init_sample((int)(numchan*floor(chunk_len*srate/1000*max_chunk_oversize_factor)),chunk);

		// and run...
		for (double endtime = lsl::local_clock()+duration;lsl::local_clock()<endtime;) {
			// run a single execution of the outlet
			{
				cout << "new outlet(" << name << "," << type << "," << numchan << "," << fmt << "," << srate << ")...";
				lsl::stream_outlet outlet(info,0,max_buffered);
				cout << "done." << endl;
				// send in bursts
				double now, start_time = lsl::local_clock(), fail_at=start_time+seconds_between_failures;
				for (int target,diff,written=0;written<max_samples && !stop_outlet;written+=diff) {
					boost::this_thread::sleep(boost::posix_time::milliseconds(chunk_len));
					now = lsl::local_clock();
					if (now>fail_at)
						break;
					target = (int)floor((now-start_time)*srate);
					int num_elements = (int)std::min((std::size_t)((target-written)*numchan),chunk.size());
					if (num_elements)
						outlet.push_chunk_multiplexed(&chunk[0],num_elements);
					diff = num_elements/numchan;
				}
			}
			cout << "del outlet(" << name << "," << type << "," << numchan << "," << fmt << "," << srate << ")" << endl;
			// downtime
			boost::this_thread::sleep(boost::posix_time::millisec(100*(rand()%50)));
		}
	} catch(std::exception &e) {
		std::cerr << "ERROR during run_outlet() Stress-test function: " <<	e.what() << std::endl;
	}
	num_outlets.fetch_sub(1);
}

    scheduled_executor default_executor()
    {
        if (!default_executor_instance.load())
        {
            scheduled_executor& default_exec =
                scheduled_executor::default_executor();
            scheduled_executor empty_exec;

            default_executor_instance.compare_exchange_strong(
                empty_exec, default_exec);
        }
        return default_executor_instance.load();
    }

            /// Acquires ownership of the \a recursive_mutex. Suspends the
            /// current HPX-thread if ownership cannot be obtained immediately.
            ///
            /// \throws Throws \a hpx#bad_parameter if an error occurs while
            ///         suspending. Throws \a hpx#yield_aborted if the mutex is
            ///         destroyed while suspended. Throws \a hpx#null_thread_id if
            ///         called outside of a HPX-thread.
            void lock()
            {
                thread_id_type const id = thread_id_from_mutex<Mutex>::call();
                HPX_ASSERT(id != thread_id_from_mutex<Mutex>::invalid_id());

                if (!try_recursive_lock(id))
                {
                    mtx.lock();
                    locking_thread_id.exchange(id);
                    util::ignore_lock(&mtx);
                    util::register_lock(this);
                    recursion_count.store(1);
                }
            }

void test_timed_executor(std::array<std::size_t, 4> expected)
{
    typedef typename hpx::traits::executor_execution_category<
            Executor
        >::type execution_category;

    HPX_TEST((std::is_same<
            hpx::parallel::execution::sequenced_execution_tag,
            execution_category
        >::value));

    count_sync.store(0);
    count_apply.store(0);
    count_sync_at.store(0);
    count_apply_at.store(0);

    Executor exec;

    test_timed_apply(exec);
    test_timed_sync(exec);
    test_timed_async(exec);

    HPX_TEST_EQ(expected[0], count_sync.load());
    HPX_TEST_EQ(expected[1], count_apply.load());
    HPX_TEST_EQ(expected[2], count_sync_at.load());
    HPX_TEST_EQ(expected[3], count_apply_at.load());
}

hpx::future<void> plain_future_void()
{
    ++count_plain_future_void;

    // make sure this function is not concurrently invoked
    HPX_TEST_EQ(count_active_plain_future_void.fetch_add(1) + 1, 1);

    hpx::this_thread::suspend(boost::chrono::microseconds(100));

    --count_active_plain_future_void;
    HPX_TEST_EQ(count_active_plain_future_void.load(), 0);

    return hpx::make_ready_future();
}

int hpx_main(int argc, char** argv_init)
{
    hpx::new_<test_server>(hpx::find_here(), moveonly()).get();
    HPX_TEST_EQ(constructed_from_moveonly.load(), 1);

    moveable o;
    hpx::new_<test_server>(hpx::find_here(), o).get();
    HPX_TEST_EQ(constructed_from_moveable.load(), 1);

    hpx::new_<test_server>(hpx::find_here(), moveable()).get();
    HPX_TEST_EQ(constructed_from_moveable.load(), 2);

    return hpx::finalize();
}

void calc_block(
    hpxla::local_matrix_view<boost::int64_t>& H 
  , hpxla::local_matrix_view<hpx::future<void> >& C // Control matrix.
  , coords start   // The start of our cell block. 
  , coords end     // The end of our cell block.
  , coords control // Our location in the control matrix.
  , std::string const& a
  , std::string const& b
    )
{
    // TODO: Handle this with hpx::wait_all?
    C(control.i,   control.j-1).get(); 
    C(control.i-1, control.j-1).get(); 
    C(control.i-1, control.j  ).get(); 

    winner local_best = H_best.load();

    // Generate scores.
    for (boost::uint32_t i = start.i; i < end.i; ++i)
    {
        for (boost::uint32_t j = start.j; j < end.j; ++j)
        {
            H(i, j) = calc_cell(i, j, a[i-1], b[j-1]
              , H(i,   j-1) // left
              , H(i-1, j-1) // diagonal 
              , H(i-1, j  ) // up
            ); 

            if (H(i, j) > local_best.value)
            {
                local_best.value = H(i, j);
                local_best.i = i;
                local_best.j = j;
            } 
        }
    }

    winner H_best_old = H_best.load(); 

    while (true)
    {
        if (local_best.value > H_best_old.value)
        {
            if (H_best.compare_exchange_weak(H_best_old, local_best))
                break;
        }
        else
            break;
    }
}