https://hgpu.org/?p=5660
Simple optimizations for an applicative array language for graphics processors