amd/common: Optimize the computation of retile maps.
Was playing around with how we store retile maps, and I wanted to remove the optimization to not calculate the map for imported textures (for another MR), and was looking at how not to regress performance. I think this MR is a good improvement even if I end up removing that optimization:
Behavior around ~1080p on a 2500U:
old:
30-60 ms on every miss
new:
5 ms initally (miss in the tile cache)
<0.5 ms afterwards
Also included a small addrlib cleanup.