Announcement

Collapse
No announcement yet.

Intel's Brewing New Linux Graphics Driver Features

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

  • Intel's Brewing New Linux Graphics Driver Features

    Phoronix: Intel's Brewing New Linux Graphics Driver Features

    Eric Anholt of Intel spoke on Saturday at FOSDEM 2012 in Belgium about the state of the Intel Linux graphics driver user-space and some of their future plans...

    http://www.phoronix.com/vr.php?view=MTA1MzE

  • #2
    Fragmentation/Isolation?

    Hi,

    I'm getting a bit worried about/for the intel camp. Based on the news I've been reading for the past year, I feel like they're isolating themselves.

    How are they isolating themselves? The main thing I see is their refusal to use Gallium3D. I admit to not knowing how complete a rewrite it would require, but getting a bunch of APIs implemented for "free" can't be something they can simply ignore, no?
    Second : TTM came, and suddenly appeared intel with GEM. I don't know the technical differences, but knowing that both radeon and nouveau use a GEMified TTM, I get a feeling something isn't quite right.
    Third : UXA/SNA. I guess an acceleration architecture tailored to specific hardware has its advantages, but it really feels like Intel is playing in a corner of the sandbox where none of the others go.
    This leads me to fragmentation : SNA/Glamor/Uxa. It sounds like a major rewrite is taking place all the time, leaving the user unsure of what is best...

    All the same, thanks to the Intel devs for their OpenGL 3.0 push, this is a great example of work that benefits everyone!

    Serafean

    Comment


    • #3
      Audio is good, but holy crap who is tapping during the whole video?

      Watched the whole video -- thanks for recording and uploading this, Michael.
      Last edited by mattst88; 02-05-2012, 12:15 PM.

      Comment


      • #4
        Originally posted by mattst88 View Post
        Audio is good, but holy crap who is tapping during the whole video?

        Watched the whole video -- thanks for recording and uploading this, Michael.
        LOL Michael, its usually a very good idea to pay attention to actually line up your video frame properly, and do an actual test record and play it back to make sure your LONG record is going to show the actual data on the board, and to check the alignment now and again, and you don't need to show almost 40% of the lower display and wall doing nothing while cutting off virtually all the data on screen, BTW doesn't your cam have a directional sock for the mike you can use so as to remove your constant tapping,typing, and beer swigging ? being picked up, none the less thanks for the effort and ill try and watch it to get the point of the missing context text.
        Last edited by popper; 02-05-2012, 02:14 PM.

        Comment


        • #5
          I wish ApiTrace were as awesome as the Linux folks said it was. I mean, the current incarnation may be great for debugging the stack, but it's not at all that useful for debugging complex applications. It's still a lot less useful in many ways than the (no longer updated, buggy, and poorly designed) gDEBugger, and of course nowhere even remotely close to PIX, PerHud, etc.

          In particular, ApiTrace can't yet trace a running application, allowing me to break, investigate source, alter memory, and continue. It really needs to be _tightly_ integrated into a traditional graphical debugger application to start being useful for application development.

          We're doing some GL ports right now (OS X, iOS; Linux only if I end up magically having a ton of freetime magically) and I wouldn't be able to get by without gDEBugger. Unfortunately it's no longer maintained (AMD has newer versions of it, but they're Windows-only and integrated in Visual Studio, and AMD totally fucked up the UI when they did it and didn't integrate the debugging features, so it's just become harder to use and more crash-prone). The older versions from Gremedy have a lot of bugs and missing features (the buffer viewer in particular is frustratingly under-featured, the stride configuration is wrong) and I get crashes a lot when using UBO's (which we're using all over the place). Plus its integrated debugger is nowhere near as nice as the many graphical Linux debuggers or Visual Studio's debugger.

          A tool like that -- with added support for debugging OpenVG, OpenCL, and ideally even D3D -- would be hardcore awesome.

          And of course bonus points if it's dead easy to integrate into an IDE like KDevelop or Anjuta or whichever is popular, so that edit and rerun cycles are as painless and quick as possible.

          Comment


          • #6
            Originally posted by phoronix View Post
            Phoronix: Intel's Brewing New Linux Graphics Driver Features

            Eric Anholt of Intel spoke on Saturday at FOSDEM 2012 in Belgium about the state of the Intel Linux graphics driver user-space and some of their future plans...

            http://www.phoronix.com/vr.php?view=MTA1MzE
            This is the first time I've heard about the gl backend to cairo since its 1.10 release a year, or so, ago.
            I really hope somethong comes of it b/c, IIRC, Untz seemed very doubtful about it (and given his benchmarks I can see why).
            Btw, I thought MSAA wasn't used often due to its overhead? I recall reading about another AA scheme that seemed to be in wider use amongst the discrete card manufacturers on AT. Using it on a low end part like Intel makes no sense, given that.

            Comment


            • #7
              Hopefully good quality videos will show up on https://www.youtube.com/user/fosdemtalks/videos

              Comment


              • #8
                Originally posted by stqn View Post
                Hopefully good quality videos will show up on https://www.youtube.com/user/fosdemtalks/videos

                well perhaps they will upload them in a months time if someone can be bothered like last year, when everyone's forgot about the event and moved on to better and newer things in the news by then.

                the most comical thing i find about all this lack of live video content in 2012 when ARM "Linux" is finally in the limelight and all the mainstream press and end users just love to cover/watch this new emerging platform/ecosystem while they wait for more windows8 PR....., is the fact this Floss event "the biggest free and non-commercial event organized by and for the community" states they will have live video of several rooms here
                http://fosdem.org/2012/news/video-st...s#comment-form and yet the links don't work, the video's if any are not kept for later re-streaming AND the "add a comment" there doesn't work "You are not authorized to view comments." LOL

                its not rocket science for any devs etc there to hock up your cam to a laptop , register and upload your video's live over the wireless network there ( i assume someone at least organised that open wireless) to http://www.dailymotion.com/relevance/search/fosdem/1 and add a FOSDEM tag etc and leave dailymotion to make the different bit-rate encodes and store them for later playback.

                or if they cant be bothered to even register than there's always the even simpler live stream option using http://veetle.com/index.php/broadcast#0 that takes all of 2 minutes or even less to start streaming the event from any reasonable i3/i5 laptop im sure many are carrying around there, it's like ALL the Floss devs don't want to capitalise on the simple video PR coverage to get more people interested and helping build linux LOL.
                Last edited by popper; 02-05-2012, 05:17 PM.

                Comment


                • #9
                  The proper multi-sampling anti-aliasing (MSAA) support for the Intel Linux driver is finally on the horizon! MSAA will become more important with the greater usage of cairo-gl.
                  Is this code going to be highly Intel specific, or could it be used to benefit other free drivers as well?

                  Comment


                  • #10
                    Originally posted by Hamish Wilson View Post
                    Is this code going to be highly Intel specific, or could it be used to benefit other free drivers as well?
                    My guess, very Intel-specific. Gallium has different paths mostly, and I think Intel is the only classic driver with hw that even can do MSAA.

                    Comment


                    • #11
                      Originally posted by stqn View Post
                      Hopefully good quality videos will show up on https://www.youtube.com/user/fosdemtalks/videos
                      FOSDEM didn't show up in the Xorg dev room at all to record. I was the only one recording any videos.

                      These videos were all just recorded from a very easy Flip Mino HD camera. Tapping? No tapping unless it was the sound of MacBook Pro keyboard or beer bottle hitting the desk...
                      Michael Larabel
                      http://www.michaellarabel.com/

                      Comment


                      • #12
                        Originally posted by elanthis View Post
                        I wish ApiTrace were as awesome as the Linux folks said it was. I mean, the current incarnation may be great for debugging the stack, but it's not at all that useful for debugging complex applications. It's still a lot less useful in many ways than the (no longer updated, buggy, and poorly designed) gDEBugger, and of course nowhere even remotely close to PIX, PerHud, etc.
                        [..]
                        It's still a pretty young project, so hopefully it will improve with time. I did however notice that icculus wrote basic profiling support for it and used that to track down a performance problem in a game so it's clearly useful for some uses.

                        Further profiling support and the planned ability to trim down traces to just a couple of frames also seems like really useful tools.

                        Comment


                        • #13
                          Originally posted by Serafean View Post
                          Hi,

                          I'm getting a bit worried about/for the intel camp. Based on the news I've been reading for the past year, I feel like they're isolating themselves.

                          How are they isolating themselves? The main thing I see is their refusal to use Gallium3D. I admit to not knowing how complete a rewrite it would require, but getting a bunch of APIs implemented for "free" can't be something they can simply ignore, no?
                          Second : TTM came, and suddenly appeared intel with GEM. I don't know the technical differences, but knowing that both radeon and nouveau use a GEMified TTM, I get a feeling something isn't quite right.
                          Third : UXA/SNA. I guess an acceleration architecture tailored to specific hardware has its advantages, but it really feels like Intel is playing in a corner of the sandbox where none of the others go.
                          This leads me to fragmentation : SNA/Glamor/Uxa. It sounds like a major rewrite is taking place all the time, leaving the user unsure of what is best...

                          All the same, thanks to the Intel devs for their OpenGL 3.0 push, this is a great example of work that benefits everyone!

                          Serafean
                          True, like the article says, Glamor sucks because it has to go through the entire OpenGL stack and that slows it down, while SNA sucks because it's back-end has to be rewritten for each hardware generation. I guess next they will "invent" a common IR between the OpenGL and SNA (and in the future OpenCL) that will allow them write a single back-end for each hardware generation. After that they will implement all the high level APIs to generate this IR. Sounds familiar? Maybe because that's exactly what Gallium3D does? But like Intel said, "there is no technical reason for Intel to switch to Gallium3D", they will just implement their own version.

                          Comment

                          Working...
                          X