Missed optimization on array construction from fixed size byte array

[tesuji]

godbolt link: https://rust.godbolt.org/z/MNuq-K

I have two snippets which I expect they have the same assembly:

use std::convert::TryInto;
pub fn foo(bytes: [u8; 16]) -> u32 {
    let mut out = [0u32; 4];
    for (a, b) in out.iter_mut().zip(bytes.chunks_exact(4)) {
        *a = u32::from_ne_bytes(b.try_into().unwrap());
    }
    out[3]
}

and

pub fn foo(bytes: [u8; 16]) -> u32 {
    let out: [u32; 4] = unsafe {
        (bytes.as_ptr() as *const u32 as *const [u32; 4]).read_unaligned()
    };
    out[3]
}

The non-optimal one has this output:

example::foo:
        sub     rsp, 24
        movups  xmm0, xmmword ptr [rdi]
        movaps  xmmword ptr [rsp], xmm0
        mov     eax, dword ptr [rsp + 12]
        add     rsp, 24
        ret

Which should be:

example::foo:
        mov     eax, dword ptr [rdi + 12]
        ret

[tesuji]

Optimized since Rust 1.52