This benchmark requires https://github.com/JuliaLang/julia/pull/5355 and building Julia with LLVM 3.4. Try flog(1000,1000,500).

gistfile1.txt

function sweep( irange, jrange, U, Vx, Vy, A, B )
    for j in jrange
        @simd for i in irange
            @inbounds begin
                u = U[i,j]
                Vx[i,j] += (A[i,j+1]+A[i,j])*(U[i,j+1]-u)
                Vy[i,j] += (A[i+1,j]+A[i,j])*(U[i+1,j]-u)
                U [i,j] = u + B[i,j]*((Vx[i,j]-Vx[i,j-1]) + (Vy[i,j]-Vy[i-1,j]))
            end
        end
    end
end

function iterate( t, U, Vx, Vy, A, B )
    m,n = size(U)
    for k=1:t
         sweep(2:m-1,2:n-1,U,Vx,Vy,A,B)
    end
end

function flog(m,n,timesteps)
    A = fill(0.2f0,m,n)
    B = fill(0.25f0,m,n)
    U = (rand(Float32,m,n)-0.5f0)*1.0f-6
    Vx = zeros(Float32,m,n)
    Vy = zeros(Float32,m,n)
    time = @elapsed(iterate(timesteps,U, Vx, Vy, A, B))
    println("GFlop = ",13.0*(m-2)*(n-2)*timesteps/time*1E-9)
end