Announcement

Collapse
No announcement yet.

Open ATI R600/700 Driver Gains Tiling Support

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

  • Open ATI R600/700 Driver Gains Tiling Support

    Phoronix: Open ATI R600/700 Driver Gains Tiling Support

    For those of you not interested in today's ATI Catalyst 10.5 for Linux driver, if you pull the very latest open-source ATI Radeon Linux graphics driver stack there is now tiling support for the R600/700 (Radeon HD 2000/3000/4000 series) graphics processors...

    http://www.phoronix.com/vr.php?view=ODI5Mw

  • #2
    I miss the point here. Is the tiling in question the same as the one done by the window managers ?

    Please Michael, add some technical background to your news for those not really into that field...

    Comment


    • #3
      Is there going to be tiling support for R5xx?!?

      Comment


      • #4
        Originally posted by Xheyther View Post
        Is the tiling in question the same as the one done by the window managers ?
        no. This is related to the internal memory layout used by textures/buffers in your video memory. Tiled buffers are faster because pixels that are close to each other are more likely to be in the same cache-line. They are more difficult to handle though, especially when you want to allow CPU access to the buffer in question.

        Comment


        • #5
          Originally posted by Xtigyro View Post
          Is there going to be tiling support for R5xx?!?
          r300g along with latest xf86-video-ati from git and pretty much any KMS kernel supports full tiling. I began to work on it in January and it became complete and bug-free at the beginning of May, and that's also the first time it got enabled on KMS by default. It wasn't top priority so it evolved slowly.

          r300c supports tiling partially.

          Comment


          • #6
            Will this boost 2d and 3d performance, or just one of them?

            Comment


            • #7
              what do you mean with 1D???

              Comment


              • #8
                The 1D and 2D names refer more to the levels of hierarchy than to dimensions (eg 1D is small tiles, 2D is bigger tiles made up of small tiles). The tiling modes we chose to use just happened to have a "D" in the name.

                Comment


                • #9
                  I tested the patches yesterday and I haven't seen any regression so far. Performance really exploded in ut2003 - previously I wasn't able to reach anything over 25fps when ramping all detail settings up (1024x768 res) but now the framerate never drops below 55fps (in fact the engine seems to throttle the fps somewhere around 60fps).

                  Nice work there!

                  Anyway, I applied the drm patches to the latest drm-linus branch.

                  Comment


                  • #10
                    Originally posted by LiquidAcid View Post
                    ... but now the framerate never drops below 55fps (in fact the engine seems to throttle the fps somewhere around 60fps).
                    Nice to hear that it's working for somebody. Isn't 60Hz the refresh rate of your monitor?

                    Comment


                    • #11
                      No, it's not a LCD if that's what you had in mind. Hooked up is an old IBM P260, running at 85Hz in all resolutions I use.

                      So VSync isn't the issue here, I think it's really some game engine setting - gonna have a look at the INIs later.

                      Comment


                      • #12
                        Just tested color-tiling with SuperTux2 on 1280x800. I observed no regressions and performance is boosted, albeit not as much as previously reported. Depending on the scene, I get an increase from 32 to 40 or from 32 to 60 fps.

                        It was really fun to see how color tiling behaves when you forget to patch mesa.

                        Comment


                        • #13
                          The initial patch said it was KMS only, When is UMS going to be supported?

                          Comment


                          • #14
                            Isn't radeon moving in the same direction as intel? KMS-only?

                            I consider UMS deprecated and I strongly suggest to use KMS, which works fairly well - even when using stable (vs. experimental) kernels.

                            Comment


                            • #15
                              Originally posted by Obscene_CNN View Post
                              The initial patch said it was KMS only, When is UMS going to be supported?
                              There are no plans to support it with UMS.

                              Comment

                              Working...
                              X