Announcement

Collapse
No announcement yet.

Weston Repainting With The New Presentation Extension

Collapse
X
 
  • Filter
  • Time
  • Show
Clear All
new posts

  • Weston Repainting With The New Presentation Extension

    Phoronix: Weston Repainting With The New Presentation Extension

    With Wayland's Weston compositor having the presentation feedback support, Pekka Paalanen of Collabora has written about Weston's repaint scheduling...

    Phoronix, Linux Hardware Reviews, Linux hardware benchmarks, Linux server benchmarks, Linux benchmarking, Desktop Linux, Linux performance, Open Source graphics, Linux How To, Ubuntu benchmarks, Ubuntu hardware, Phoronix Test Suite

  • #2
    Pekka is very productive at what he's doing. Good job.

    Comment


    • #3
      how about
      per application worst time in last n refreshes with misses treated extra
      and compositors worst time added to it

      it's all integers, so it would be light on the cpu

      i think exposing next vblank time would be useful to heavy real time applications
      as they can choose to render again if there is time to (instead of triple buffering)

      Comment


      • #4
        Originally posted by gens View Post
        how about
        per application worst time in last n refreshes with misses treated extra
        and compositors worst time added to it

        it's all integers, so it would be light on the cpu

        i think exposing next vblank time would be useful to heavy real time applications
        as they can choose to render again if there is time to (instead of triple buffering)
        Making use of variable sync is the way forward.
        Many laptops already have the hardware.

        Comment


        • #5
          Originally posted by liam View Post
          Making use of variable sync is the way forward.
          Many laptops already have the hardware.
          yes
          and this has nothing to do with it

          Comment


          • #6
            Originally posted by gens View Post
            how about
            per application worst time in last n refreshes with misses treated extra
            and compositors worst time added to it

            it's all integers, so it would be light on the cpu

            i think exposing next vblank time would be useful to heavy real time applications
            as they can choose to render again if there is time to (instead of triple buffering)
            In the software world there are no minimum time guarantees.
            Your minimum time guarantees for the compositor is infinite time!
            You build such algorithms starting with the assumption the static GUI and application will make it in time.
            This is an assumption you threat it as an assumption, not a certainty.
            If the timing is not met you wait until the render is ready and then update the render to the screen.

            You don't start out assuming/knowing when something be finished. Doing this is useless on modern multitasking computer systems.
            Last edited by plonoma; 15 February 2015, 09:29 AM.

            Comment


            • #7
              Originally posted by plonoma View Post
              In the software world there are no minimum time guarantees.
              Your minimum time guarantees for the compositor is infinite time!
              You build such algorithms starting with the assumption the static GUI and application will make it in time.
              This is an assumption you threat it as an assumption, not a certainty.
              If the timing is not met you wait until the render is ready and then update the render to the screen.

              You don't start out assuming/knowing when something be finished. Doing this is useless on modern multitasking computer systems.
              you probably won't read this

              not minimum time, maximum
              it's not infinite and can be predicted
              usually vsync is the deadline so in this case; time to vsync - worst_case_compositor_overhead

              knowing the deadline is important to game engines (or at least having a consistent timing, that is achieved with this)
              with full screen opengl under X there is no compositor overhead and every flip returns right after it's done

              Comment

              Working...
              X