Re: exchange_and_add inlined in g++ 4.0.2

[Paolo Carlini <pcarlini at suse dot de>]

bert hubert wrote:

>>No, this is incorrect. In the *.so there is the i486+ version, only,
>>it's not inlined.
>>    
>>
>Ok - does the .so contain only one implementation depending on the
>./configure of g++, or does it autodetect?
>
Only one, which would be the i486+ version these days...

> Is my understanding correct that
>when configured w/o any architecture options on "regular intel" platforms
>the i386 version gets compiled in?
>  
>
No. Just have a look to your configure.host: there you can see clearly
that for i486-*, i586-*, i686-* the i486-* is chosen.

>>>It is the top function that is related to the pdns_recursor in this system-wide profile:
>>>      
>>>
>>Well, certainly I should know the application in some detail in order to
>>be able to tell you something meaningful, which is not the case.
>>However, in general, I see numbers below 2%...
>>    
>>
>This is system wide, including kernel and idle cycles. Relative to my
>application is is more like 15%, which is real enough.
>  
>
Indeed. I would appreciate more details about your application, in that
case.

>>>Will do. I think I can provide my own __exchange_and_add as an inline?
>>>      
>>>
>>Sure! If you are willing to poke a bit into the sources (this is free
>>software, after all! ;) you can very easily fix the problem to suit your
>>    
>>
>More exactly I meant, if I add an inlined __gnu_cxx::__exchange_and_add to
>my binary, would that work, w/o recompling libstdc++ etc.
>  
>
I don't think you can do much without recompiling libstdc++-v3, if you
want something clean and not dirty tricks... Consider that inside the
library there are also instantiations of string and wstring.

>>needs. Thanks for filing the PR.
>>    
>>
>Doing that now, thanks.
>  
>
Thank you,
Paolo.