Support for non-temporal instructions

[pfotouhi]

Currently, we do not have good support for non-temporal instructions and we are basically ignoring the non-caching behavior and implement them as normal (cacheable) instructions (see MOVNTI, MOVNTQ, MOVNTDQ, MOVNTS, MOVNTD). This implementation seems to be functionally correct, but have impact on performance with possibly some corner case issues.

We have the flag "uncacheable" in microop, but we need to have:

• Proper mechanisms in Ruby/Classic to deal with this (make sure non-caching is maintained).

• Have a micro-op to flush the cacheline (if it exist in cache, options suggested by Gabe: clflushopt and clwb).

P.S. Looking at some x86 stuff, we have cdaf (which we don't implement) which is basically cda (check data address) + a cache line eviction. If we have this one implemented we can simply use that instead of cda we have there.