On Tue, Jan 12, 2016 at 05:28:24PM +0200, Mika Kuoppala wrote: > The most common thing on normal operation is ring tail > pointer update. Put it first in the shadow register list for > gen8, like we do with gen9.
With execlists it is not. Also BSD2? Do you have the complete list of shadowed regs available? NEEDS_FORCE_WAKE() is just reg<40000, so I am inclined to favor that test first. Do you have a couple of profiles to show how wrong am I? Note that the first course of action would probably to special case the hot writes as I915_WRITE_FW. -Chris -- Chris Wilson, Intel Open Source Technology Centre _______________________________________________ Intel-gfx mailing list Intel-gfx@lists.freedesktop.org http://lists.freedesktop.org/mailman/listinfo/intel-gfx
