`am*_db.x` vs `am*.x + dbar` for acquire or release memory orders

[xry111]

For example: __atomic_exchange_n(p, v, __ATOMIC_ACQUIRE);

Is it better to use amswap.w then dbar 0b10100 instead of amswap_db.w?

Cc @heiher

[heiher]

We did a performance comparison in JVM, and there was almost no difference on LA664. This is not meant to apply to other cases.

[xry111]

Ok, so continue to using amswap_db for now as it's shorter.