Add support for prefetching.

[edanor]

Prefetch operations are operations that inform the CPU that a given fragment of memory will be used soon. Memory fragment subject for prefetching should be moved into cache on a preferred memory level, selected by the user.

While prefetch operations don't have too much to do with vectors (although they are exposed using intrinsics), it might be useful to add some vector-specific operations together with prefetch operations.

The proposal is to add PREFETCH0, PREFETCH1 and PREFETCH2 instructions to vector interface.

[Note to users] Please add comments if you are interested in this feature and/or if you have any hints on this idea.

[edanor]

Interface now supports PREFETCH0, PREFETCH1 and PREFETCH2 instructions. For Intel extensions these should map to _mm_prefetch(...).