Smart paging to video cards or clusters using OpenCL

I've been working on setting up a 'smart paging' library for my OpenCL projects. Basically this involves checking:

• The number of de开发者_如何学编程vices
• Sizes of each one's memory tiers
• The largest power of two sized buffer that will fit into all devices of a group
• The largest power of two sized buffer that will communicate between host and devices
• The largest power of two sized buffer will fit into the biggest device(s).
• Various other things to use the left over memory asynchronously
• Various other things too general to list

These results determine aspects of the page size, timing, etc.. It is just what I came up with off the top of my head.

Are there any useful paradigms, architectures or other conventions which already accomplish efficient paging OR are there other properties I should be focused on or are any of these inadequate/irrelevant?

I think this multi-plaform library should solve your idea:

https://github.com/emrainey/OpenCL-Environment