Lower BinOp::Cmp to llvm.{s,u}cmp.* intrinsics

[DaniPopes]

Lowers mir::BinOp::Cmp (three_way_compare intrinsic) to the corresponding LLVM llvm.{s,u}cmp.i8.* intrinsics.

These are the intrinsics mentioned in #118310, which are now available in LLVM 19.

I couldn't find any follow-up PRs/discussions about this, please let me know if I missed something.

r? @scottmcm

[nikic]

LLVM 19 doesn't have all the optimization support for these intrinsics yet.

[DaniPopes]

Should the check be on LLVM 20?

[scottmcm]

I'm excited to do this, but if nikic says it's too early, that probably means it shouldn't be on by default yet.

Maybe make this PR adding a -Z flag for it, so people can play with it and test out how close it is?

(Then in the future once it's fully ready we can remove that flag again)

[nikic]

Should the check be on LLVM 20?

Yes, that's fine. You'll have to test locally against LLVM 20 though to make sure codegen tests don't break.

[DaniPopes]

@rustbot ready

I didn't add the unstable flag since LLVM 20 already tries to lower the manual icmp pattern to the intrinsic so doing it ourselves shouldn't change much.

[scottmcm]

Given that https://rust-lang.zulipchat.com/#narrow/channel/187780-t-compiler.2Fwg-llvm/topic/LLVM.2020/near/495752627 is talking about landing LLVM 20 in nightly in just a couple weeks, should we wait for that before doing this PR? I don't know how much we get from having it now in LLVM 19, when the LLVM 20 versions of the tests won't run in bors.