With memory_order_conservative, the semantics is full bidirectional fencing.
One full fence before the atomic operation, and one after. But for some
reason, with fetch_and_add, add_and_fetch and xchg, there is only a trailing fence, and there is no leading fence at all. This is a violation of our memory ordering rules.
One full fence before the atomic operation, and one after. But for some
reason, with fetch_and_add, add_and_fetch and xchg, there is only a trailing fence, and there is no leading fence at all. This is a violation of our memory ordering rules.
- relates to
-
JDK-8293547 Add relaxed add_and_fetch for macos aarch64 atomics
-
- Resolved
-