Announcement

Collapse
No announcement yet.

MLAA For Mesa Is Ready For Testing

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

  • #21
    Originally posted by Drago View Post
    So...this MLAA is entirely in software/shader, and not actual GPU anti-aliasing method?
    Yeah. But it's entirely on the GPU, no CPU software part.

    Comment


    • #22
      Irrlicht Engine version 1.7.2
      Linux 2.6.39-ARCH #1 SMP PREEMPT Mon Jun 27 21:26:22 CEST 2011 x86_64
      Using renderer: OpenGL 4.1.10834
      ATI Radeon HD 5700 Series: ATI Technologies Inc.
      OpenGL driver version is 1.2 or better.
      GLSL version: 4.1
      Loaded texture: AreaMap33
      Unsupported texture format
      Quit message received.

      Ran 204.882s, average fps 225.12, min 1

      Average on fps 1055.90, average off fps 181.66

      MLAA took on average 0.9ms

      mesa
      Version : 7.10.99.git20110706-1
      Radeon 5770

      Comment


      • #23
        I'm also getting some really odd numbers with the latest version of the test, and this is using the binaries you provided.

        Comment


        • #24
          Originally posted by smitty3268 View Post
          You're running the software renderer there, instead of the r300g driver.
          Ops, good catch. Hopefully that explains the really broken display too. So next up is diagnosing why it didn't go with r300g. Sorry for all the noise.

          Comment


          • #25
            Okay, trying again. It's funny how llvmpipe gave me correct numbers with the previous commit; let's see if the latest improves things on real hw.

            Comment


            • #26
              Yeah, the new results seems to be okay:

              Code:
              Ran 41.635s, average fps 288.01, min 1
              
              Average on fps 130.86, average off fps 583.15
              (The first test I ran was with power_profile set to low, so quite far off the mark, sorry about that!)

              Comment


              • #27
                No, it was interesting to see how much the low profile affected the performance. Almost exactly four times worse with low.

                Comment


                • #28
                  Originally posted by curaga View Post
                  Window position anomaly - maybe caused by your compositor?

                  Timing wrong - possibly. Could also be that you use a newer irrlicht, as the title bar fps is taken from irr.
                  You were right, it was KWins fault. The average FPS is now also correct with the new version.

                  Results with r600g and hd5770 (juniper):

                  low: Average on fps 60.43, average off fps 312.62
                  high: Average on fps 60.38, average off fps 526.58

                  with MLAA on, the FPS was somehow capped at 60FPS. That's probably because SwappbuffersWait is turned on, but why isn't the FPS capped without MLAA?

                  Comment


                  • #29
                    The test app sets vblank_mode to 0 to avoid vsync. Swapbufferswait can't be overridden from the app though.

                    Comment


                    • #30
                      Originally posted by curaga View Post
                      The test app sets vblank_mode to 0 to avoid vsync. Swapbufferswait can't be overridden from the app though.
                      Yeah, I was wondering why it didn't cause the sphere to run at 60FPS without MLAA.

                      Results without SwapbuffersWait:
                      low: Average on fps 86.40, average off fps 811.24
                      high: Average on fps 162.36, average off fps 1540.74

                      Comment

                      Working...
                      X