Compiler fails to remove jumps using jump table if every label in it is same

[AngelicosPhosphoros]

I tried this code:

pub enum K{
    A(Box<[i32]>),
    B(Box<[u8]>),
    C(Box<[String]>),
    D(Box<[u16]>),
}

pub fn get_len(arg: &K)->usize{
    match arg {
        K::A(ref lst)=>lst.len(),
        K::B(ref lst)=>lst.len(),
        K::C(ref lst)=>lst.len(),
        K::D(ref lst)=>lst.len(),
    }
}

I expected to see this happen:

Since layout of each enum variant is same, I expected to get machine code without any jumps, e.g.

example::get_len:
        mov     rax, qword ptr [rdi + 16]
        ret

Instead, this happened:
Compiler generates jump table, every branch of it points to the same code, and code jumps using it despite being it redundant.

example::get_len:
        mov     rax, qword ptr [rdi]
        lea     rcx, [rip + .LJTI0_0]
        movsxd  rax, dword ptr [rcx + 4*rax]
        add     rax, rcx
        jmp     rax
.LBB0_1:
        mov     rax, qword ptr [rdi + 16]
        ret
.LJTI0_0:                                     ; Note that each of it jumps to same label at the end.
        .long   .LBB0_1-.LJTI0_0
        .long   .LBB0_1-.LJTI0_0
        .long   .LBB0_1-.LJTI0_0
        .long   .LBB0_1-.LJTI0_0

Meta

rustc --version --verbose:

rustc 1.68.0 (2c8cc3432 2023-03-06)
binary: rustc
commit-hash: 2c8cc343237b8f7d5a3c3703e3a87f2eb2c54a74
commit-date: 2023-03-06
host: x86_64-unknown-linux-gnu
release: 1.68.0
LLVM version: 15.0.6
Compiler returned: 0

Also, godbolt link.

[taiki-e]

This seems a regression introduced in Rust 1.65: https://godbolt.org/z/7vnT1z67E

@rustbot label +regression-from-stable-to-stable

[apiraino]

WG-prioritization assigning priority (Zulip discussion).

@rustbot label -I-prioritize +P-medium +T-compiler +I-slow

[nikic]

It's another typed GEP issue.

[AngelicosPhosphoros]

llvm/llvm-project#62570

[erikdesjardins]

Fixed in 1.73, presumably due to the LLVM 17 upgrade (#114048): https://godbolt.org/z/WTcKvcfh4

This specific example seems to be a backend optimization (the optimized IR is the same in 1.72 and 1.73); the general case is fixed in nightly by #121665: https://godbolt.org/z/cxnsWoWzx.

@rustbot label +E-needs-test