Skip to content

mesa: don't count buffer references for the context that created them

Marek Olšák requested to merge mareko/mesa:buffer-no-refcount into master

If this is applied on top of all my other stuff, performance increases by 10% in one viewperf subtest.

We know that atomics are slow on both x86 and ARM. The major impact of this is that glBindBuffer now doesn't use atomics, which may be called even more often than draw calls.

Merge request reports