Update benchmark results for `wasmi`

[vshymanskyy]

According to https://wasmi-labs.github.io/blog/posts/wasmi-v0.32 wasmi-v0.32 should display much better results

It would be nice to see the updated numbers. Cheers!

@ejpbruel2 @Robbepop

[Robbepop]

@vshymanskyy I just updated the benchmarked Wasm runtimes to:

• Wasmi v0.38.0
• Wasmtime v25.0.2
• And a special fork to make Wasm3 work again (wasm3-rs is broken unfortunately)

I benchmarked using Macbook M2 Pro: (Note that Wasmi is known to run badly on Mac.)

Here are the results:

Runtime	Coremark
stitch	3035
wasm3	2928
wasmi	1701
wasmtime	12721

Note: Wasmi v0.31 had a score of 788. Therefore: >100% improvement

I used the following optimization profile:

[profile.release]
opt-level = 3
lto = "fat"
codegen-units = 1

I'd say it is a tie between Wasm3 and Stitch. Wasmi is still using a simple switch-loop and hopefully soon also using tail-calls for its dispatch. 🤞

edit: With the release of Wasmi v0.39 I re-benchmarked the runtimes on M2 Pro: (Wasmi v0.39 is ~8% faster than v0.38 and ~130% faster than v0.31.)

Runtime	Coremark
stitch	3017
wasm3	2851
wasmi	1830
wasmtime	12732