https://svn.boost.org/trac10/ticket/6830 <p> I created a simple benchmark for measuing raw allocation throughput for 3 classes of different sizes with a common base class (constructors and destructors trivial). The number of allocations was set to 40,000,000 as it was roughly giving me 10 seconds running time per test. </p> <p> it turns out that on VC++9 (release target with default optimizations) boost::make_shared is significantly slower than simply doing boost::shared_ptr(new). Here's the benchmark output: </p> <p> <a class="missing wiki">TestBoostMakeShared</a> 10.577s 3.78179e+006 allocs/s </p> <p> <a class="missing wiki">TestBoostSharedPtrNew</a> 8.907s 4.49085e+006 allocs/s </p> <p> As you can see boost::make_shared is over 15% slower than boost::shared_ptr(new) idiom. </p> <p> One suggested solution: </p> <p> boost::shared_ptr doesn't have a way to retrieve the deleter without using RTTI which is what is slowing down the execution on VC++9/10. I decided to add one and use it from an alternative boost::make_shared. So I did the following: </p> <ol><li>I added a virtual function to detail::sp_counted_base (detail\sp_counted_base_w32.hpp): </li></ol><blockquote> <p> virtual void * get_raw_deleter( ) = 0; </p> </blockquote> <ol start="2"><li>I implemented get_raw_deleter() function in sp_counted_impl_p (detail\sp_counted_impl.hpp): </li></ol><blockquote> <p> virtual void * get_raw_deleter( ) { </p> <blockquote> <p> return 0; </p> </blockquote> <p> } </p> </blockquote> <ol start="3"><li>I implemented get_raw_deleter() function in sp_counted_impl_pd (detail\sp_counted_impl.hpp): </li></ol><blockquote> <p> virtual void * get_raw_deleter( ) { </p> <blockquote> <p> return &amp;reinterpret_cast&lt;char&amp;&gt;( del ); </p> </blockquote> <p> } </p> </blockquote> <ol start="4"><li>I implemented get_raw_deleter() function in sp_counted_impl_pda (detail\sp_counted_impl.hpp): </li></ol><blockquote> <p> virtual void * get_raw_deleter( ) { </p> <blockquote> <p> return &amp;reinterpret_cast&lt;char&amp;&gt;( d_ ); </p> </blockquote> <p> } </p> </blockquote> <ol start="5"><li>I added the following function to detail::shared_count: </li></ol><blockquote> <p> void * get_raw_deleter( ) const { </p> <blockquote> <p> return pi_? pi_-&gt;get_raw_deleter( ): 0; </p> </blockquote> <p> } </p> </blockquote> <ol start="6"><li>I added the following function to shared_ptr&lt;&gt;: </li></ol><blockquote> <p> void * _internal_get_raw_deleter( ) const { </p> <blockquote> <p> return pn.get_raw_deleter( ); </p> </blockquote> <p> } </p> </blockquote> <ol start="7"><li>I made a separate copy of boost::make_shared function and replaced a single line from: </li></ol><blockquote> <p> boost::detail::sp_ms_deleter&lt; T &gt; * pd = boost::get_deleter&lt; boost::detail::sp_ms_deleter&lt; T &gt; &gt;( pt ); </p> </blockquote> <p> to: </p> <blockquote> <p> boost::detail::sp_ms_deleter&lt; T &gt; * pd = static_cast&lt;boost::detail::sp_ms_deleter&lt; T &gt; *&gt;(pt._internal_get_raw_deleter()); </p> </blockquote> <p> Benchmarking the results afterwards gave me the following results on VC++9: </p> <p> <a class="missing wiki">TestBoostSharedPtrNew</a> 9.204s 4.34594e+006 allocs/s </p> <p> <a class="missing wiki">TestBoostMakeShared</a> 10.499s 3.80989e+006 allocs/s </p> <p> <a class="missing wiki">TestBoostMakeSharedAlt</a> 7.831s 5.1079e+006 allocs/s </p> <p> These changes translated into almost 35% improvement in allocation speed over the current implementation of boost::make_shared. Or to put it differently, they amount to 25+% decrease in running time as we could have supposed from the profiling results. </p> en-us /htdocs/site/boost.png https://svn.boost.org/trac10/ticket/6830 Trac 1.4.3 Dave Abrahams Thu, 26 Apr 2012 05:38:57 GMT https://svn.boost.org/trac10/ticket/6830#comment:6 https://svn.boost.org/trac10/ticket/6830#comment:6 <ul> <li><strong>status</strong> <span class="trac-field-old">new</span> → <span class="trac-field-new">closed</span> </li> <li><strong>resolution</strong> → <span class="trac-field-new">fixed</span> </li> </ul> <p> (In <a class="changeset" href="https://svn.boost.org/trac10/changeset/81899" title="Merged revision(s) 81860-81861 from trunk: Change make_shared to use ...">[81899]</a>) Merged revision(s) 81860-81861 from trunk: Change make_shared to use the new _internal_get_untyped_deleter. Fixes <a class="closed ticket" href="https://svn.boost.org/trac10/ticket/6830" title="#6830: Patches: make_shared slower than shared_ptr(new) on VC++9 and 10 (closed: fixed)">#6830</a>. ........ Add allocate_shared_noinit. ........ </p> Ticket