Trouble with Float32

[oschulz]

With

function jselfdotavx(a)
    s = zero(eltype(a))
    @inbounds @avx for i ∈ eachindex(a)
        s += a[i] * a[i]
    end
    s
end

This works:

jselfdotavx(rand(Float64, 1024))

but with Float32

jselfdotavx(rand(Float32, 1024))

I get

ERROR: MethodError: no method matching +(::VectorizationBase.SVec{4,Float32}, ::VectorizationBase.SVec{4,Float64})
[...]

[chriselrod]

Thanks. Your example should now work on the latest release, but there is still some "trouble with Float32", so I'll leave this issue open until at least:

• No more broadcast issues with Float32.
• You can define x_i = zero(eltype(x)) within a for loop, instead of needing to define x_i = z, where z is set equal to zero outside of the loops. That is:

function mygemmavx!(C, A, B)
    z = zero(eltype(C))
    @avx for i ∈ 1:size(A,1), j ∈ 1:size(B,2)
        Cᵢⱼ = z
        for k ∈ 1:size(A,2)
            Cᵢⱼ += A[i,k] * B[k,j]
        end
        C[i,j] = Cᵢⱼ
    end
end

Is unnecessarily awkward to write.

If any new issues crop up, I can add them to that list.

[chriselrod]

I fixed both those remaining problems. So if/when new issues crop up, they should get their own (new) issue.

[oschulz]

Thanks for the super-quick fix, @chriselrod!