Host fallback

[ProGTX]

The specification requires host fallback when there is no OpenCL device available. I see some options how this may be implemented, however I believe it would take a lot of work.

What needs to be done:

• Partitioning of work-groups
  • Work-item synchronization
• Computing inside data reference
  • Would first require Static type checking of device types #2 to be implemented
  • This computation does not always have to occur, but how to achieve that?
• Other hidden features ...