See https://discourse.julialang.org/t/gpu-genetic-algorithm/112700/11. Since the async form has the steps for the PSO stepping as part of a compilable kernel itself, it is possible to do an entire PSO as a GPU kernel. There are some applications that want to run a global optimization over many parameters, such as the one described in that Discourse post, which could use such an approach. And given the code we have here, I think it wouldn't be too hard to make that accessible.
See https://discourse.julialang.org/t/gpu-genetic-algorithm/112700/11. Since the async form has the steps for the PSO stepping as part of a compilable kernel itself, it is possible to do an entire PSO as a GPU kernel. There are some applications that want to run a global optimization over many parameters, such as the one described in that Discourse post, which could use such an approach. And given the code we have here, I think it wouldn't be too hard to make that accessible.