When benchmarking our upgrade to wasmtime 34 we found a regression when it comes to accessing wasm memory in the host. Here's a repo with a some benchmarks with a Wasm module that calls a host function in a loop: https://github.com/adambratschikaye/wasmtime-memory-regression

When the host function accesses the Wasm memory there's a ~30% slowdown when switching from wasmtime 33.0.1 to 34.0.1 (just run cargo bench with both versions to see the difference). When the host function doesn't do anything the performance is unchanged.

Going through a git bisect on the wasmtime repo, it seemed that the regression was coming from #10877.

Versions and Environment

Operating system: Ubuntu 24.04

Architecture: x64