Announcement

Collapse
No announcement yet.

Mesa Threading Support Is Slow, Still Being Developed

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

  • Mesa Threading Support Is Slow, Still Being Developed

    Phoronix: Mesa Threading Support Is Slow, Still Being Developed

    One of the ways that Intel has been trying to make their Mesa driver faster is through proper threading support, but for now the support is unfortunately slower while the code is still being actively developed...

    http://www.phoronix.com/vr.php?view=MTI5NTg

  • #2
    The buffering of calls seems like something the applications could do themselves.
    Not sure how much gains I'm supposed to expect with this.
    (Depends heavily what calls and how, when the application makes them. Instancing anyone?)

    It would be neat if OpenGL could have that you stash the calls on the gpu and can call a single call to execute a bunch of them. Somewhat like some sort of a function call.

    Comment


    • #3
      Originally posted by plonoma View Post
      The buffering of calls seems like something the applications could do themselves.
      Not sure how much gains I'm supposed to expect with this.
      (Depends heavily what calls and how, when the application makes them. Instancing anyone?)

      It would be neat if OpenGL could have that you stash the calls on the gpu and can call a single call to execute a bunch of them. Somewhat like some sort of a function call.
      Given how many function calls it takes to do stuff in GL, it can be a win. Even for drawing, much better to let another thread do the kernel and hardware calls than to add latency to the app itself, boyh for GL and other APIs. Break the core driver into another process a la the more stable and secure WDDM and this adds another layer of win.

      Comment


      • #4
        Originally posted by elanthis View Post
        Break the core driver into another process a la the more stable and secure WDDM and this adds another layer of win.
        That already exists; LIBGL_ALWAYS_INDIRECT / __GL_FORCE_INDIRECT.

        Comment

        Working...
        X