Casting to u64 directly is equivalent to first casting to i64, which does
sign extension. For negative values, this fills the highest 32 bits with
ones rather than the zeros we need.