Mesa 21.0.2 Released With Lavapipe Fixes, Improved AMD L3 Cache Calculation
Accumulating for Mesa 21.0.2 is the usual random smattering of fixes but with no area dominating the change-log this time around. Mesa 21.0.2 has just a few fixes for the likes of the Radeon and Intel drivers but nothing too exciting there. The other changes include several Lavapipe Vulkan CPU driver fixes, disabling of sparse buffers on GFX7/GFX8 for RadeonSi, Mesa state tracker fixes, and a few EGL fixes too.
One somewhat notable change is a rework to the AMD CPU L3 cache affinity code to correct the Zen L3 cache pinning code previously added to Mesa. This came as a fix since the existing CPU detection broke on 128-core AMD (dual EPYC) systems. The new AMD L3 cache affinity code works better now for those large multi-CPU systems and will also gracefully deal with big/little CPU types should they appear in the future for AMD processors. David Airlie explained in the original patch, "It now iterates over all CPU cores, gets the core count for each CPU, and works out the L3_ID from the physical CPU ID, and the current cores L3 cache. It then tracks how many L3 caches it has seen and reallocate the affinity masks for each one."
The list of patches to this minor Mesa 21.0.2 release can be found via the release announcement.
9 Comments
