[boost] [shared_ptr] sp_counted_base_cw_x86.hpp vs. the rest.

14 Apr 2005

      Rene Rivera wrote:
...
I running the timing tests again, with some changes to see if I can get 
more representative numbers. But attached is the end result.
With a changed test that exercises all three functions, here's the 
timing results:

sp_counted_base_w32.hpp...
debug: 15.673 - release: 7.151

sp_counted_base_cw_x86.hpp...
debug: 15.943 - release: 6.499

sp_counted_base_nt.hpp...
debug: 15.933 - release: 6.499

So the the hand assembly is as fast as the non-threading version. But is 
both slower and faster than the Interlocked* version. What gives? Is 
there a better smart_ptr timing test out there? I'm including the one I 
used.

-- 
-- Grafik - Don't Assume Anything
-- Redshift Software, Inc. - http://redshift-software.com
-- rrivera/acm.org - grafik/redshift-software.com
-- 102708583/icq - grafikrobot/aim - Grafik/jabber.org

#include <boost/config.hpp>

#if defined(BOOST_MSVC)
#pragma warning(disable: 4786)  // identifier truncated in debug info
#pragma warning(disable: 4710)  // function not inlined
#pragma warning(disable: 4711)  // function selected for automatic inline expansion
#pragma warning(disable: 4514)  // unreferenced inline removed
#endif

//
//  shared_ptr_timing_test.cpp - use to evaluate the impact of thread safety
//
//  Copyright (c) 2002 Peter Dimov and Multi Media Ltd.
//
// Distributed under the Boost Software License, Version 1.0. (See
// accompanying file LICENSE_1_0.txt or copy at
// http://www.boost.org/LICENSE_1_0.txt)
//

#include <boost/shared_ptr.hpp>
#include <boost/weak_ptr.hpp>
#include <iostream>
#include <vector>
#include <ctime>

int const n = 4 * 1024 * 1024;
int const r = 4;

int main()
{
    using namespace std;

    std::vector< boost::shared_ptr<int> > v(n);
    std::vector< boost::weak_ptr<int> > w(n);
    std::vector< boost::shared_ptr<int> > x(n);
    boost::shared_ptr<int> pi(new int);

    clock_t t = clock();
    {
        for (int i = 0; i < r; ++i)
        {
            v.assign(n,pi);
            w.assign(v.begin(),v.end());
            for (int j = 0; j < n; ++j) { x[j] = w[j].lock(); }
            v.clear();
            w.clear();
            x.clear();
        }
    }
    t = clock() - t;

    std::cout << static_cast<double>(t) / CLOCKS_PER_SEC << '\n';

    return 0;
}