[cpp-threads] Memorymodel counterproposal:synchronization operations

[Nelson, Clark]

> Sorry for not in turn being clear. I'd like to know
> which operations you have in mind. For example,
> Is a read that is guaranteed to not be re-ordered
> with previous reads a syncop?

Well, I would say that any operation that imposes sequencing on accesses
to non-volatile objects is a synchronization operation. (A sequence
point imposes sequencing, but is not an operation. And of course all
volatile accesses are sequenced, unless they aren't separated by a
sequence point.)

But my whole point is that, in the language of the standard, sequencing
constraints on accesses to (non-volatile) X should not be imposed by
accesses to (any sort of) Y. There may of course be library routines
that access an object and also force synchronization, but those are
semantically separate operations.

Clark