Removal of aarch64 workaround for float fns?

[kolektiv]

This commit: 278eb28 introduced a workaround for functions special-casing aarch64 due to an upstream LLVM issue. That issue was closed some time ago, but (according to the comment) that couldn't be removed all the time that rustc supported LLVM versions which wouldn't contain that fix. As rustc now requires LLVM 21 (which is post-fix, I think), could that special-casing now be removed?

Happy to try and help where possible, though I'm no LLVM expert. I have used the intrinsics directly without issue, although "works on my machine" is clearly not the standard.

[programmerjake]

This was fixed in LLVM 19, Rust currently requires LLVM 20

[kolektiv]

This was fixed in LLVM 19, Rust currently requires LLVM 20

Ah, I stand corrected! 👍 Still, shouldn't be an LLVM problem anymore - and it doesn't seem to have caused any regressions at least as far as test results go.

[theemathas]

It's LLVM 21 as of Rust 1.91.0 https://doc.rust-lang.org/releases.html#internal-changes

[kolektiv]

I think this can be closed as #497 has been merged - can always be resurrected if needed!

🥳

[programmerjake]

It's LLVM 21 as of Rust 1.91.0 https://doc.rust-lang.org/releases.html#internal-changes

that's the default LLVM version, not the minimum which is LLVM 20 as of Rust 1.92.0