[Strs2FillMyDrms]

As LL/SC type architectures get more and more widespread, I am wondering all the novel ways in which programmers may discover new "anomalies" with the old memory model rules and constraints. speculative execution combined with relaxed semantics... and now LL/SC widespread bubbleless cache exclusiveness acquirement?

In all honestly mathematical operations are still better performed by LOCK prefixed architectures since there is no reason to go to the assembler level to re-read the cache... if all you're going to be doing is perform a mathematical operation.

LL/SC is good for displays/publishing operations (losers win all) ... not for graphics or mathematical ops.