Difference between `zero(eltype(x))` and `zero(x)` in Julia

In ‘Julia High Performance’ (2nd Ed.), the author gives the following example to explain how to obtain type stability in loops:

function simdsum_fixed(x)
    s = zero(eltype(x))
    @simd for v in A
        s += v
    end
    return s
end

In a previous example, with different code they simply use zero(x). I’ve therefore tried the same function without eltype() as below (simdsum_fixed_b). They seem to work in the same way.

function simdsum_fixed_b(x)
    s = zero(x)
    @simd for v in A
        s += v
    end
    return s
end

How is then zero(eltype(x)) different from zero(x)? What is the advantage of using the first over the second?

Thanks in advance.

>Solution :

Use zero(eltype(x)) if x is a container (e.g. a vector) and you want additive identity of its elements. Use zero(x) if you want to get the additive identity of x itself.

Example with vector as here the difference is most clearly visible:

julia> x = [1, 2, 3]
3-element Vector{Int64}:
 1
 2
 3

julia> zero(x)
3-element Vector{Int64}:
 0
 0
 0

julia> zero(eltype(x))
0