[cpp-threads] Revised Atomic Paper

[Lawrence Crowl]

On 8/3/07, Nelson, Clark <clark.nelson at intel.com> wrote:
> There's an HTML nesting problem in the "Prior Approaches" section, such
> that the entire rest of the document is indented a few levels.

Fixed.  It is in part a problem with Internet Explorer not paying
attention to proper tag nesting.

> This is just a heads-up: in considering this and the memory model
> proposal simultaneously, it is obvious to me that we need to define the
> term "read-modify-write operation" somewhere in 1.10. I will have some
> words for that in this mailing. You might want to delete the phrase "in
> the sense of ... in 1.10p7" in the four places it occurs. At the very
> least, you should delete the paragraph number from the reference; in the
> standard, the granularity of a reference is a section.

Right.  I purposely held off that change until the formal wording.  In
the meantime, please think about whether we need more subsections
for finer reference.

> Also, I don't think you need to add the parenthesized phrase "(when not
> memory_order_relaxed)" in the discussions of various RMW operations. The
> definition of "synchronizes with" already distinguishes relaxed
> operations from others.

Okay.

>
> I understand the motivation for defining a global variable to provide
> global synchronization. But given that users will in any event have to
> modify their programs, and will have to insert the name of a global
> variable to get global synchronization, is there any practical benefit
> to using one provided by the standard over one provided by the user?

Yes, the various libraries need to agree on a name, and the standard
makes that possible.  There will be a paper in the mailing on this
topic.  You may have it by now.

-- 
Lawrence Crowl