abrown closed issue #2153:
PR #2149 provides an initial implementation of wasm atomics with x64/newBE. That includes an initial lowering of CLIF
AtomicRmwinto a multi-instruction sequence that is general, but inefficient in the case where the old value at the location is not required, which is probably a common occurrence. In such cases we could do much better by producing a single LOCK-prefixed integer read-modify-write instruction, for examplelock xorq $27, (amode).
Last updated: Dec 06 2025 at 06:05 UTC