Those of you at the office yesterday may remember me getting very excited about perf events suggesting gtt thrashing on DOTA2.
On further testing, i915:i915_gem_evict says that we aren't evicting things from the GTT after the first run. We still get the i915:i915_gem_request_wait_begin stalls on the subdatas, which I believe are because the BOs are new, and we're doing the first faulting, which in the kernel assumes synced behavior. Swapping our unsynced maps to CPU maps (I think this is safe, and doesn't have the same faulting leading to stalling on the GPU problem) didn't improve performance, so I'm fairly sure that we can move on from this issue at this point.
pgpBwPxK4FUU0.pgp
Description: PGP signature
_______________________________________________ mesa-dev mailing list mesa-dev@lists.freedesktop.org http://lists.freedesktop.org/mailman/listinfo/mesa-dev
