Using atomic operations right? I got a way to do it without them, or memory
barriers... Its heavily distributed per-thread reference counting.
Per-thread proxy counting... E.g., more than one object can share the same
proxy counter... The PDR scan logic takes this into account by iterating a
generation of deferred objects against the per-thread counters... A very
simple hash on the pointer value gets to the counter. The counter is can
also be in per-object metadata... It allows a thread to grab multiple
references to an arbitrary number of nodes. The references are also
copyable. This is done through an indirection of the proxy counters...
Also, I should mention another form of reference counting I invented here:
What do you think of the reference counting algorithm I created for my
P.S. -- This allocator is one of the fastest and most scaleable memory
allocators out there... It was included in my initial round vZOOM
submission... It outperforms Hoard by a wide margin... It outperforms these
algorithms by a wide margin:
Luckily, the patent application for this is underway... I had to do it. This
memory allocator is $FUXKING blazing fast!