Technically something like this is relevant, but most people would assume that in practice it is going to be way too much work to get it implemented on GPUs and in such a way that it doesn't add too much overhead.