Make UB transmutes really UB in LLVM

[scottmcm]

Ralf suggested in #143410 (comment) that UB transmutes shouldn't be trapping, which happened for the one path that PR was changing, but there's another path as well, so this PR changes that other path to match.

r? codegen

[rustbot]

Some changes occurred in compiler/rustc_codegen_ssa

cc @WaffleLapkin

[RalfJung]

@nikic is this the best way to indicate "unreachable" to LLVM in the middle of a block? Does LLVM quickly normalize this by making the entire block unreachable or can such an assume cause problems?

[dianqk]

@nikic is this the best way to indicate "unreachable" to LLVM in the middle of a block? Does LLVM quickly normalize this by making the entire block unreachable or can such an assume cause problems?

According to https://llvm.org/docs/Frontend/PerformanceTips.html#id11, this should be store i1 true, ptr poison, align 1.

[RalfJung]

Ah! We should have a helper method for this then.

[scottmcm]

Oh, interesting! I'll switch it (both here and in the other spot).

@rustbot author

[dianqk]

Ah! We should have a helper method for this then.

Yes. This makes sence to me.
BTW, this reminds me of #127740.

[nikic]

FWIW I don't think it particularly matters how Rust emits this.

[scottmcm]

Well, if nothing else it's reasonable for there to be a method for "I wanted unreachable but not a terminator", so I can do that and if it ends up not mattering which code pattern it is, then fine.

I do think here is different from #127740 because there it was a "runtime" value being passed to the assert, not a constant being created to be passed to it, so there I think it's better to let LLVM normalize, but here where I never specifically wanted an assume in the first place I don't mind just emitting using a different pattern.

[scottmcm]

@rustbot ready

[WaffleLapkin]

@bors r+

[bors]

📌 Commit f5fc872 has been approved by WaffleLapkin

It is now in the queue for this repository.