www.digitalmars.com         C & C++   DMDScript  

digitalmars.D - A move semantics benchmark

reply bearophile <bearophileHUGS lycos.com> writes:
A link found through Reddit, a benchmark about Move Semantics Benchmark in C++
STL:
http://cpp-next.com/archive/2010/10/howards-stl-move-semantics-benchmark/

Bye,
bearophile
Oct 14 2010
next sibling parent so <so so.do> writes:
Wow it looks great! Too bad you have to wait so many years to get an  
improvement like move.

On Thu, 14 Oct 2010 15:29:52 +0300, bearophile <bearophileHUGS lycos.com>  
wrote:


 A link found through Reddit, a benchmark about Move Semantics Benchmark  
 in C++ STL:
 http://cpp-next.com/archive/2010/10/howards-stl-move-semantics-benchmark/

 Bye,
 bearophile




-- 
Using Opera's revolutionary e-mail client: http://www.opera.com/mail/
Oct 14 2010
prev sibling parent reply Peter Alexander <peter.alexander.au gmail.com> writes:
== Quote from bearophile (bearophileHUGS lycos.com)'s article

 A link found through Reddit, a benchmark about Move Semantics




 http://cpp-next.com/archive/2010/10/howards-stl-move-semantics-




 Bye,
 bearophile



I don't really like that benchmark. I think it will deceive people
into thinking that supported move semantics are going to give your
programs huge speed improvements. In reality, you wouldn't do all
those things with those massive objects.

A real benchmark would take some standard C++ program, add any
extra move constructors where applicable, and then compare the
performance to the program without move constructors. That way,
you're actually seeing what move semantics will really bring to
the performance of a real program.
Oct 14 2010
parent reply bearophile <bearophileHUGS lycos.com> writes:
Peter Alexander:


 That way, you're actually seeing what move semantics will
 really bring to the performance of a real program.



There are synthetic benchmarks, and various shades of more realistic
benchmarks. Both kinds have advantages and disadvantages, and both have their
place and may be useful. That's a synthetic benchmark.

Bye,
bearophile
Oct 14 2010
parent Peter Alexander <peter.alexander.au gmail.com> writes:
On 14/10/10 7:30 PM, bearophile wrote:

 Peter Alexander:


 That way, you're actually seeing what move semantics will
 really bring to the performance of a real program.



 There are synthetic benchmarks, and various shades of more realistic
benchmarks. Both kinds have advantages and disadvantages, and both have their
place and may be useful. That's a synthetic benchmark.

 Bye,
 bearophile



Fair enough, although I find this benchmark pretty useless. All it says 
is that move semantics can provide performance improvements in contrived 
situations. You can show demonstrate just about anything in contrived 
situations.
Oct 14 2010