Is there a way to use ability to branch to speed up raytracing? Can I create with Parallella raytracer in some way more efficient than gpu based ones?