NVPTX: No "undefined reference" error is raised when it should be

[japaric]

STR

$ cargo new --lib kernel && cd $_

$ edit src/lib.rs && cat $_

#![no_std]

extern {
    fn bar();
}

fn foo() {
    unsafe {
        bar()
    }
}

$ cat nvptx64-nvidia-cuda.json

{
  "arch": "nvptx64",
  "cpu": "sm_20",
  "data-layout": "e-i64:64-v16:16-v32:32-n16:32:64",
  "llvm-target": "nvptx64-nvidia-cuda",
  "max-atomic-width": 0,
  "os": "cuda",
  "panic-strategy": "abort",
  "target-endian": "little",
  "target-pointer-width": "64"
}

$ edit Cargo.toml && cat $_

[profile.dev]
debug = false  # cf. rust-lang/rust#38785

$ cargo install xargo --vers 0.3.0 || true

$ xargo rustc --target nvptx64-nvidia-cuda -- --emit=asm

$ cat $(find target/nvptx64-nvidia-cuda/debug -name '*.s')
.version 3.2
.target sm_20
.address_size 64

.extern .func bar
()
;

.func _ZN6kernel3foo17h9be095784726d4d0E()
{


        bra.uni         LBB0_1;
LBB0_1:
        { // callseq 0
        .reg .b32 temp_param_reg;
        call.uni
        bar,
        (
        );
        } // callseq 0
        bra.uni         LBB0_2;
LBB0_2:
        ret;
}

Note that bar is called by foo but it's not defined in the PTX module. Trying to load this PTX module, with cuModuleLoadData, generates a runtime error: CUDA_ERROR_NO_BINARY_FOR_GPU.

Expected behavior

Other targets that produce an executable would fail to compile this with an "undefined reference" error which is actually raised by the linker. The NVPTX targets don't involve a linker but they shouldn't produce incomplete PTX modules like the one above; instead the should raise an error like the other targets do.