Potential miscompile on Wasm / Chrome when Simd128 is enabled about parry

[kettle11]

I have created a repository with a minimal reproducable example: https://github.com/kettle11/min_repro_chrome_simd_bug/tree/main

[kettle11]

I opened a Chrome bug: https://bugs.chromium.org/p/chromium/issues/detail?id=1313647

[sebcrozet]

Thank you for investigating this! It is interesting to see that the issue happens even if you haven’t enabled any of the simd-related features (simd-stable or simd-nightly) of parry. So, could the auto-vectorisation performed by the Rust compiler be the one that messes up?

What happens if you try with the feature simd-nightly enabled (in which cases the SIMD instructions generated should depend on the what the packed_simd crate generates)?

[kettle11]

What happens if you try with the feature simd-nightly enabled (in which cases the SIMD instructions generated should depend on the what the packed_simd crate generates)?

Good idea! I just checked and it still happens even if simd_nightly is enabled.

However in the process of testing that I noticed the issue only occurs after the first page refresh (if the dev console is closed). That was not the case in my larger project. Another weird detail!

[kettle11]

Fixed in an upcoming Chrome release!

https://bugs.chromium.org/p/chromium/issues/detail?id=1313647#c22