If I need more precision that float supports, what other options do I have in OpenCL? From what I see double precision may not be available to all platforms.