Re:Secrets of dispatch_once; x86 performance trickery
by rossjudson on Saturday June 07, 2014 @09:42AM
(#7862054)
This feels racy, and I'm not sure it will work on multi-node NUMA architectures. It's one thing to delay and hope that your write is visible in the local processor cache; it's another to ensure that write is visible to other processor nodes, when there's no explicit cache coherence rule that says it will be.
[ Parent
| on HN ]
There are no comments on this comment yet.