Merged requested to merge bnieuwenhuizen/mesa:vram-cs into master
- +3% in basemark
- +1 fps in SotTR (on a base of 46, so ~2% on face value but no clue how rounding works, so could be 0-4%. Both the pre and post numbers are very consistent though, returning exactly the same number for each of 3 runs)
Tested CPU performance with bnieuwenhuizen/vulkan_tests@180c2b67 with no difference found above the noise floor.
Why do this with full VRAM visibility: less CPU-visible VRAM pressure lowers the risk for triggering moving buffers around. Why not do this on APUs: very high VRAM pressure overall