[jabl]

Sure, since it's a stronger model, so code which works on weaker acquire/release hw will work on TSO hw as well. You might as well say that sequential consistency maps very efficiently to a acquire/release model too in the same way.

[gpderetta]

Isn't TSO the closest practical implementation to a acquire/release model? What are the practical differences?

I know that TSO allows more easily to recover sequential consistency with additional barriers (Intel strengthened their original memory model to TSO for this reason).

[jabl]

No, acquire/release is a much weaker model than TSO. TSO is basically sequential consistency (SEQCST) + a store buffer. I.e. stores don't go immediately to main memory, but rather via a store buffer. Loads first peek into the store buffer of the local CPU core before going to memory. The practical effect is that in contrast to SEQCST stores may be reordered after loads ("store->load" reordering more formally).

Acquire-release consistency allows many more reorderings in addition to store->load (load->load, load->store, store->store).

For more info see e.g. Table 5 in http://www.rdrop.com/users/paulmck/scalability/paper/whymb.2...