intel: Set gfx20 async compute tuning registers and fix tuning value for gfx125
What does this MR do and why?
We were missing the async compute tuning for gfx20 platforms and had a different than recommended value for gfx125. Also Iris did not had any async compute tuning, what is also covered here.
I still don't have any performance numbers with this MR, will get it next week.