Re: notes on current atomics config and usage

[Paolo Carlini <pcarlini at suse dot de>]

Mark Mitchell wrote:

> Yes.  Assuming you want to support both 386 and 486+ in libstdc++ (and I
> think you should), we need (a) a feature macro (defined by the compiler,
> or by libstdc++ at configure-time) to see if the builtins are available,
> (b) libstdc++ code to test that feature macro.
>
> Then, the compiler default is just a detail; if the user is building a
> multilib with -march=i386, then you get the slow way; if they're
> building with -march=i486, then you get the fast way.
Ok.

As a first step in this large project, when I will be back from a few 
days of vacations, I will start on implementing __GCC_ATOMIC_BUILTINS: 
if I remember correctly should not be a very big deal, I understand you 
are in favor of adding the facility, and, last but not least, even alone 
would be of great help for code not calling into the library, like, for 
example, the lockfree shared_ptr currently in the making. In such cases 
there are no interoperability issues between user code and library code 
and we can win very easily!

Paolo.