On Sat, Jan 21, 2017 at 09:25:26AM +0000, Chris Wilson wrote: > In order to maximise concurrency between engines, if we queue a request > to a current idle ring, reorder its dependencies to execute that request > as early as possible and ideally improve occupancy of multiple engines > simultaneously.
This is more of a toy, but I'm interested in seeing how it fares nevertheless. There may be a large delay between scheduling and submitting (depending on how quick its dependencies run, and whether it has any external deps) and so querying hw state at scheduling time is optimistic. Furthermore it can be easily exploited to prioritise bad clients. Still I think the actual impact will be very small, and so am interested in the experiment on real machines - where it will be dwarfed by display boosting. -Chris -- Chris Wilson, Intel Open Source Technology Centre _______________________________________________ Intel-gfx mailing list Intel-gfx@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/intel-gfx
