Merged requested to merge mareko/mesa:zen-thread-affinity into master
- New OS-agnostic thread affinity and getcpu functions including Windows support (@jrfonseca for awareness)
- The L3 cache topology code was wrong, so I rewrote it from scratch to make Mesa on my AMD CPU faster. Hopefully it's right this time.
- It could also be used on Xeons or any multi-CPU system if anybody cares.
- I completely changed the L3 cache pinning optimization for glthread to be more effective. The old one didn't do much for me.