Unexpected NaN result

[Moelf]

julia> using LoopVectorization

julia> function f(rounds)
           pi = 1.0

           @turbo for i in 2:(rounds + 2)
               x = (-1)^iseven(i)
               pi += x / (2 * i - 1)
           end

           return pi*4
       end

julia> f(10)
NaN

julia> f(10) #simd is okay with this and sees a speed up compared to plain loop
3.058402765927333

[chriselrod]

Minimal reproducer:

julia> using VectorizationBase, LoopVectorization

julia> LoopVectorization.pow_fast(-1, iseven(MM{8}(3)))
Vec{8, Float64}<NaN, NaN, NaN, NaN, NaN, NaN, NaN, NaN>

julia> (-1) ^ iseven(MM{8}(3))
Vec{8, Float64}<1.0, -1.0, 1.0, -1.0, 1.0, -1.0, 1.0, -1.0>