Supporting weaker "relaxed" memory orderings

[rmahdav]

Currently, the Relaxed Atomics proposal focuses on introducing acquire-release memory ordering. While this is a significant improvement over sequential consistency, there are use cases that could benefit from even weaker orderings, such as C++'s relaxed ordering or load-store ordered atomics.

We should discuss the feasibility and design of supporting "more relaxed" orderings in this proposal.
Previous discussion: WebAssembly/shared-everything-threads#60 (comment)

[jakobkummerow]

Semi-related: this proposal's name is currently misleading :)

Established definitions of memory orderings (e.g. C++, Rust) tend to distinguish between "relaxed" and "acquire-release". So while acqrel is undoubtedly more relaxed than sequentially-consistent, it's still not the same as "relaxed".

So if this proposal doesn't decide to add actual "relaxed" memory orderings too (on which I don't have much of an opinion), then it should be renamed to "acquire-release atomics" to avoid confusion, and to leave the door open for a potential future proposal that does add relaxed atomics.