@cmuratori As long as the destination is aligned anyway. An atomic add "locks" no more than an atomic store does.
-
-
Replying to @cmuratori
@rygorous The LOCK prefix was an actual system bus lock. Later, they made it so that it didn't have to lock if the core had it in cache.2 replies 0 retweets 0 likes -
Replying to @cmuratori
@rygorous But I believe it _still_ took the system lock if it wasn't in the cache after that (P4 and up). Not sure about _now_.1 reply 0 retweets 0 likes -
Replying to @cmuratori
@rygorous But point being, it _was_ a real lock, and I assume that it still is a real lock on _some_ processors out there.2 replies 0 retweets 0 likes -
Replying to @cmuratori
@cmuratori No, the system bus lock is not something anyone has used for aligned operands ever since x86 cores started implementing MESI.2 replies 0 retweets 0 likes -
Replying to @rygorous
@cmuratori Not that it matters for the definition of wait-freedom; even if you do have to shut down everyone else on the bus,1 reply 0 retweets 0 likes -
Replying to @rygorous
@cmuratori that might be a slow operation, but it's still bounded-time and guaranteed progress as long as a single core can't hog the bus.1 reply 0 retweets 0 likes -
Replying to @cmuratori
@cmuratori My point is that an atomic RMW in a write-back cached setting does not require any machinery you don't already need for a write.2 replies 1 retweet 0 likes
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.