stdcxx-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Martin Sebor <>
Subject Re: Benchmarking stdcxx for Windows
Date Thu, 06 Apr 2006 13:38:35 GMT
Craig Chariton wrote:

> I took Andrew Black's posting of February 10, 2006 concerning
> benchmarking
> stdcxx and applied it to Windows.  I decided to benchmark stdcxx 4.1.3
> prior
> to the modifications made following Andrew's initial tests.

Thanks for doing that. Unfortunately, due to the nature of Andrew's
benchmark program many of the results are almost entirely determined
by the lack of one (controversial) optimization that we didn't
implement in 4.1.3 (because of the C++ Standard issue(s) referenced
in the bug report tracking the performance problem in Jira).

> I will then
> modify the library with the updates designed to resolve some of the
> performance issues and run these tests again to determine what effect,
> if
> any, the modifications have on performance. 

These numbers will be of interest if only to confirm that the
optimization has taken care of the performance issue when using
MSVC on Windows as well. But even more helpful than rerunning the
same set of benchmarks would be creating a more focused benchmark
that exercises individual member functions without the overlap in
the original code. I.e., a benchmark that produces one result for
each member function of the tested class.


View raw message