Quoting Chris Wilson (2018-01-23 17:05:57) > We only use the preempt context to inject an idle point into execlists. > We never need to reference its logical state, so tell the GPU never to > load it or save it. > > Suggested-by: Daniele Ceraolo Spurio <daniele.ceraolospu...@intel.com> > Signed-off-by: Chris Wilson <ch...@chris-wilson.co.uk> > Cc: Michal Winiarski <michal.winiar...@intel.com> > Cc: Michel Thierry <michel.thie...@intel.com> > Cc: Michal Wajdeczko <michal.wajdec...@intel.com> > Cc: Tvrtko Ursulin <tvrtko.ursu...@intel.com> > Cc: Mika Kuoppala <mika.kuopp...@intel.com> > --- > Please check the register definition. I know the save-inhibit bit used > to exist, I don't know if it is BIT(1) in RING_CONTEXT_CONTROL -- it > feels like it should be, and the numbers indicate that it does > something beneficial.
Fwiw, seems to be within normal variation. Just had a string of bad runs on drm-tip contrasting with a set of good runs with the patch. With a large enough sample size, no significance. :( Maybe with a bit more tuning this will have larger impact. Still, I think it's the right thing to do. -Chris _______________________________________________ Intel-gfx mailing list Intel-gfx@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/intel-gfx
