4 ms·
It's mostly not the atomic op overhead that's expensive, though -- it's the sharing. You can write shared-nothing algorithms using ARC objects that are local t
by loeg 5y ago
It's mostly not the atomic op overhead that's expensive, though -- it's the sharing.
You can write shared-nothing algorithms using ARC objects that are local to a single thread or core, and while it will be slightly more expensive than RC objects, the N^2 effect of sharing between N cores won't occur.