Quoting Tvrtko Ursulin (2019-02-04 12:37:00) > > On 04/02/2019 12:28, Chris Wilson wrote: > > Quoting Tvrtko Ursulin (2019-02-04 12:14:38) > >> > >> On 04/02/2019 08:41, Chris Wilson wrote: > >>> Expose whether or not we support the PMU software tracking in our > >>> scheduler capabilities, so userspace can query at runtime. > >> > >> I am leaning towards thinking PMU is a backend and not the scheduler > >> feature. We could export it via engine discovery for instance. > > > > The sw metrics are buggy. They include semaphore time on top of busy, > > but historically that has always been separate (and how it's measured by > > the HW). > > Time to resurrect the LRCA context runtime patches and see if that is > consistent in wait vs poll mode. > > But, why are the semantics of busy time related to the question of > whether to expose this flag at engine or scheduler level? The accuracy (and meaning) presented to the user currently depends on internal details that are not exposed. I just piggy backed caps.scheduler as it was adjacent to the code and already being used to determine implementation details from igt / mesa. -Chris _______________________________________________ Intel-gfx mailing list Intel-gfx@xxxxxxxxxxxxxxxxxxxxx https://lists.freedesktop.org/mailman/listinfo/intel-gfx
