On Thu, Aug 25, 2011 at 12:13 AM, Stephan T. Lavavej <stl@exchange.microsoft.com> wrote:
[Gottlob Frege]
1. I wish that C++11 atomics defaulted to acquire-on-read and release-on-write, and that's how I will almost always use them. It is also what java and C# use, I believe (on volatiles).
That doesn't solve the "independent reads, independent writes" problem, which is why the non-Standard semantics that VC8 gave volatile aren't really useful. (volatile: a multithreaded programmer's worst enemy.)
What's the "independent reads, independent writes" problem? I've probably heard, seen, or mistakenly had a bug due to it once, but don't know it by name.
The Standard's focus on sequential consistency by default was carefully thought out.
I find that most lockfree development uses acquire/release, not SC, so if the focus was to standardize existing patterns (at least in lockfree), then in that sense it seems odd. But I do believe it was carefully thought out, and by people who know better than I. I have a slight worry that SC was chosen so that it was easier to understand in general, instead of for those who will probably use it most, but again, I wasn't there, and really, I think that the memory ordering of an atomic operation is important enough that I might argue that it should always be specified, so I don't at all mind that I need to be explicit to use acqure/release. (sorry for the run on sentence) Tony