Announcement

Collapse
No announcement yet.

Future of ATI Linux drivers

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

  • #11
    Oh, it gets complicated. It depends on whether you're talking about "the 2d engine" or "the 2d API calls".

    The 2d *engine* doesn't change from earlier chips to 5xx, and on the 6xx the 2d functions are pretty much the same although they're emulated in microcode.

    The 2d APIs are a different story -- I believe XAA can be done on the 2d engine but EXA needs the 3d engine for most of the really useful API calls. I think you can sort-of do rotation on the 2d engine but the 3d engine handles it much better. Blending absolutely needs the 3d engine and blending (the core of XRender) seems to be what EXA is really all about.

    There is also general agreement that EXA needs a TTM-style memory manager to really shine, but that's another story.
    Last edited by bridgman; 14 February 2008, 12:54 PM.
    Test signature

    Comment


    • #12
      Interesting.

      Well in that case, it looks like full and proper 3D support is the key to 2D performance on the new cards. The order of priorities in driver development make very good sense in that context.

      Thanks for the information bridgman, and for all your hard work! Tell the devs Happy Valentine's, from all of us here.

      Comment


      • #13
        Hmm. At least with Nvidia moving 2d operations to 3d units dropped the performance considerably, every user is complaining how their new 8800 gt is way slower with Firefox and desktop in general than their old 7950 ultra..

        So, how is the situation with AMD currently? Are the 2d operations slower, equal, or faster with latest (3d) hardware compared to the ones with dedicated 2d units?

        Comment


        • #14
          Originally posted by Porter View Post
          Wow, that's interesting. Is doing 2D in shaders faster than in overlay hardware? What's the benefit?
          More generic functionality. If you've got stream processors that can do that sort of stuff reasonably well, it's better overall to pile a few more of them into a piece of hardware that will only get used to scale 2D video playback.

          Comment


          • #15
            Originally posted by curaga View Post
            Hmm. At least with Nvidia moving 2d operations to 3d units dropped the performance considerably, every user is complaining how their new 8800 gt is way slower with Firefox and desktop in general than their old 7950 ultra..

            So, how is the situation with AMD currently? Are the 2d operations slower, equal, or faster with latest (3d) hardware compared to the ones with dedicated 2d units?
            That has less to do with the act and more to do with the implementation, I suspect.

            Comment


            • #16
              Originally posted by bridgman View Post
              The biggest part of our Linux market is workstation, which realistically needs closed drivers today and in the future.
              Not trying to joust with you here, John- care to enlighten us as to WHY they need to be closed?

              They don't need to have DRM playback.

              They don't need anything special that I'm aware of, actually, save the immediate mode pre-processor to accelerate the old-style immediate mode code that some of the workstation vendors insisted upon keeping around.

              I can see possibly only that piece there. What else have I missed?

              Comment


              • #17
                Originally posted by bridgman View Post
                Just a reminder that xv support on any 5xx or higher part needs 3d first, since the video processing is done in shaders rather than in the overlay hardware. It's not a question of "just turning on the scaler" any more.
                Personally ... i wouldn't care if they implement it in software and emulate the xv interface. That would help a lot of people. Just as a filler of course..

                Todays CPUs (btw. a AMD 5600+ for me are fast enough for that stuff and can handle scaling pretty easy. Considering peoples problems with cpu load(scrolling) i am sure that most of these problems depend on implementations in software for some functions. Using the current radeonHD driver gives me a perfect video using the VLC software renderer (720p scaled to 1050p).

                Of course that just a guess but there are two possibilities:

                a) bridgeman thinks "he is too close to the truth..."

                or

                b) bridgman is using a handkerchief wiping his tears away.. and mumbling "if he knew..."

                well of course i don't want do insult anyone - nor the development team nor bridgeman. It is just a joke and i hope that it's clear.

                cleric

                Comment


                • #18
                  Originally posted by Svartalf View Post
                  I can see possibly only that piece there. What else have I missed?
                  Optimization?

                  Comment


                  • #19
                    Originally posted by Svartalf View Post
                    Not trying to joust with you here, John- care to enlighten us as to WHY they need to be closed? They don't need to have DRM playback. They don't need anything special that I'm aware of, actually, save the immediate mode pre-processor to accelerate the old-style immediate mode code that some of the workstation vendors insisted upon keeping around. I can see possibly only that piece there. What else have I missed?
                    I think you've pretty much got it. The main point is that it takes a lot of investment to make a fully featured, high performance workstation driver (including a lot of optimization work) and no company feels like giving that investment away and making things easier for competitors.

                    Originally posted by cleric View Post
                    Personally ... i wouldn't care if they implement it in software and emulate the xv interface. That would help a lot of people. Just as a filler of course..
                    Do you think that would be useful ? I don't think it was something we ever considered. Is there a benefit to having software scaling via Xv, maybe some app that only works with Xv and not with another API ? I *think* we'd have to write the scaling and CSC code anyways, just in x86 rather than shaders, wouldn't we ?
                    Last edited by bridgman; 14 February 2008, 03:26 PM.
                    Test signature

                    Comment


                    • #20
                      Hey John, care to answer me too?

                      Comment

                      Working...
                      X