On Mon, 26 Feb 2018, Chris Wilson <chris@xxxxxxxxxxxxxxxxxx> wrote: > Quoting Jani Nikula (2018-02-26 14:00:37) >> On Thu, 22 Feb 2018, Tvrtko Ursulin <tursulin@xxxxxxxxxxx> wrote: >> > From: Tvrtko Ursulin <tvrtko.ursulin@xxxxxxxxx> >> > >> > Introduce subplatform mask to eliminate throughout the code devid checking >> > sprinkle, mostly courtesy of IS_*_UL[TX] macros. >> > >> > Subplatform mask initialization is moved either to static tables (Ironlake >> > M) or runtime device info init (Pineview, Haswell, Broadwell, Skylake, >> > Kabylake, Coffeelake and Cannonlake). >> >> I thought Chris had the goal of separating runtime and static init, and >> I very much agreed with that idea. Throw away the mkwrite stuff. This >> patch seems to be at odds with that goal by tying a runtime init into >> the same mask with statically initialized platform mask. > > Yes. > > In the extreme version of single platform LTO, we would bake one > device-info stanza for every subplatform. That may be a little overkill > (or rather too complicated for the user to know and too fine-grained to > be useful). So the middle ground is that we have subplatform in the > runtime_info, and the compiler has to do two loads. We can still benefit > from using BIT() though. Well, let's not conflate platform and subplatform into the same mask in the same info for the short-term benefit, then? And with that, I think we could go towards: #define IS_BWD_ULT(dev_priv) (IS_BROADWELL(dev_priv) && IS_ULT(dev_priv)) where the IS_ULT would just check the subplatform mask. I am not convinced it's worth overloading the subplatform bits. BR, Jani. -- Jani Nikula, Intel Open Source Technology Center _______________________________________________ Intel-gfx mailing list Intel-gfx@xxxxxxxxxxxxxxxxxxxxx https://lists.freedesktop.org/mailman/listinfo/intel-gfx
