[llvm-branch-commits] [llvm] release/22.x: [ARM] Fix inlining issue in ARM (#169337) (PR #177974)

[Nikita Popov via llvm-branch-commits]

nikic wrote:

> So not exactly correctness, but so much performance that it effectively equates to correctness?

Yes, basically.

Some related issues: https://github.com/rust-lang/rust/issues/131745 (can't use get_unchecked) https://github.com/rust-lang/stdarch/pull/1964 (can't use vreinterpret)

This *seems* like a relatively safe backport to me because it just extends the feature whitelist (which I believe is unique to ARM, I don't think any other target handles these via a whitelist and assumes superset compatibility by default). On the other hand, the issue is not exactly new, so waiting another release isn't the end of the world either.

https://github.com/llvm/llvm-project/pull/177974