Re: [Boost-users] boost::function performance issues with C++11

Subject: Re: [Boost-users] boost::function performance issues with C++11
From: ·¹Í° (athrun911500_at_[hidden])
Date: 2014-07-08 04:59:41

sure, I will report this to GCC community.

At 2014-07-08 02:58:48, "Antony Polukhin" <antoshkka_at_[hidden]> wrote:

2014-07-08 4:28 GMT+04:00 ·¹Í° <athrun911500_at_[hidden]>:

Hi, there,
Thanks for you guys help. I think I have reached the answer I wanted. So I'd like to explain why there are performance issues with C++11 in boost::function.
First of all, I wanna reply the guys who give me hints.

>> What if you try with just 1 integer? Does the difference in time stay proportionally the same?
There won't be big differences if there is just 1 integer. So this proofs that it could be a memory copy problem.

>> I'd run the example code through Callgrind (valgrind --tool=callgrind) 
>> and then compare the reports in KCachegrind. This should at least give 
>> you an idea of the source of the problem.
You are right. I ran callgrind, and it showed some interesting things. If I compiled without C++11, it shows the most hot spot is "wordcopy_fwd_align". Otherwise, the most hotspot is "boost::bind" and there is no records for wordcopy_fwd_align. I think that means if we dont' use C++11, there will be some memory copy optimizations.

>> Maybe trying another compiler would help you obtain your "proof".
>> How about intel compiler? 
>> It also has the "-std=c++11" flag, so, if it is a compiler issue it will be clear.
As you suggested, I tried Intel compiler. It's great. The data shows there is no performance differences between with or without C++11 in boost::function. That proofs your guess that it's a compiler issue.

So here is my conclusion, there won't be memory copy optimization when using (C++11, boost::function, gcc4.9/clang3.4). But Intel compiler does well to do C++11 related optimizations.

Could you please create a ticket in GCC's bugzilla for this issue? Post the most interesting parts of this discussion in that ticket and then post a link to the ticket in this mailing list.

Thanks in advance!

--
Best regards,
Antony Polukhin