Paul A. Clayton (paaronclayton.delete@this.gmail.com) on November 21, 2012 9:39 am wrote:
> bakaneko (nyan.delete@this.hyan.wan) on November 21, 2012 6:52 am wrote:
> [snip]
> > This way you don't need to add superfluous reads to protect
> > objects which span multiple cachelines, which is as annoying
> > as multiple unrelated objects in one cacheline.
>
> When would one need to guard a value that is not accessed as part of the critical
> section? Is such an artifact of certain languages (e.g., providing object-granular
> locks) or are there cases where such provides a useful guard?

Good question, I just assumed it is, but
yeah, you are right, it's probably something
which happens rarely and not a good idea.

Even as guard, it would just make other
transactions fail more often...