iter::zip changes vectorization behaviour even if value is unused

I tried this code:

godbolt

#[inline(never)]
pub fn zipped(
    slice_of_refs: &mut [&mut bool],
    unused_slice: &mut [u8],
) {
        for (unused, write_destination) in unused_slice
        .iter_mut()
        .zip(slice_of_refs) {
            **write_destination = true;
        }
}


#[inline(never)]
pub fn not_zipped(
    slice_of_refs: &mut [&mut bool],
    unused_slice: &mut [u8],
) {
    for write_destination in slice_of_refs {
        **write_destination = true;
    }
}

I expected to see this happen: They compile to (approximately) the same assembly (Either both vectorized or both scalar).

Instead, this happened: Only the code which zips (but does not use) another slice is vectorized. (Caveat: so far I could only observe this with -target-cpu=alderlake set, and I am unsure whether vectorized or scalar operations would be considered the 'correct' result here)

Meta

rustc --version --verbose:

Full compiler version
rustc 1.87.0 (17067e9ac 2025-05-09)
binary: rustc
commit-hash: 17067e9ac6d7ecb70e50f92c1944e545188d2359
commit-date: 2025-05-09
host: x86_64-unknown-linux-gnu
release: 1.87.0
LLVM version: 20.1.1
Internal compiler ID: r1870

Reproduce the zipped and not_zipped functions from the issue in Godbolt with rustc 1.87.0, -C opt-level=3, and -C target-cpu=alderlake, then compare their generated assembly. No repository file or test is named; done means identifying why the unused zipped value changes vectorization and adding a focused regression test or fix in the relevant compiler area.