Broadcasting in Kernels

using DiffEqGPU, OrdinaryDiffEq
pa = [1.0]
u0 = [3.0]

function f(du,u,p,t)
    du[1] = 1.01 * u[1] * p[1]
end

function f2(du,u,p,t)
    du .= 1.01 .* u .* p
end

prob = ODEProblem(f2, u0, (0.0, 1.0), pa)

function prob_func(prob, i, repeat)
  remake(prob, u0 = 0.5 .+ i/100 .* prob.u0)
end

ensemble_prob = EnsembleProblem(prob, prob_func = prob_func)
sim = solve(ensemble_prob, Tsit5(), EnsembleGPUArray(), saveat = 0.1, trajectories = 100)

Start by running the Julia example with EnsembleGPUArray and compare the f and f2 definitions, focusing on the broadcasted operations in f2. Trace how solve handles this kernel path and establish whether the example should run successfully; done means the intended broadcasting behavior is reproducible and verified for the shown ODE problem.