[RFC] libstdc++ atomicity, x86 agnostic version

[Paolo Carlini]

Benjamin Kosnik wrote:

>I expect some comments about the fact that these atomic functions will
>no longer be inlined, impacting performance. Therefore, I attach the a
>performance testrun with the patch, which shows a minimal impact.
>
This seems really sad and more specific tests should be performed: the 
current
performance testsuite definitely is not tuned to benchmark that.

It's well known that changing the kind of refcounting/locking affects 
**heavily**
the performance of a refcounted basic_string implementation.

Indeed, I'm working on a patch allowing to build a non-refcounted 
basic_string
but, on the other hand, there are *many* applications (in particular, 
most of
the non-MT) which currently benefit enormously from the efficient refcounted
scheme. Recently, I have discovered at least two places in the library 
were we
didn't properly share the _Rep, that is we weren't fully exploting the full
potential of our refcounted implementation.

I really would like to have 2 different possibilities: non-refcounted,
aggressive, top-notch refcounted, this should make everyone happy.

I'm afraid that non inlining such a strategic piece of code may have 
**very**
bad effects...

Let's try to work together on alternative solutions...

Paolo.