LLVM horizontal reduction intrinsics broken on AArch64

[gnzlbg]

These include:

• floating-point sum and product
• integer wrapping_sum and wrapping_product
• horizontal min/max reductions: min_element/max_element
• bitwise horizontal reductions: and/or/xor
• mask reductions all,any, none

Reported to LLVM: https://bugs.llvm.org/show_bug.cgi?id=36796

[nikic]

This issue has been fixed upstream.

[gnzlbg]

@nikic has LLVM been updated to a version with the fix?

[nikic]

@gnzlbg Nope. The relevant monorepo commit is llvm/llvm-project@aa7cfa7#diff-4cd9888950373f65c4898b144777d044. Right now we're still on the LLVM 8 branch (unsure if that means that we will no longer track LLVM trunk?)

[SparrowLii]

@nikic It looks like it has been fixed? As it passed the tests normally on my aarch64 machine.

[nikic]

@SparrowLii Yes, the relevant LLVM upgrade has happened a long time ago by now :)