No announcement yet.

The State Of ARM SoC GPU Linux Drivers

  • Filter
  • Time
  • Show
Clear All
new posts

  • #11
    Originally posted by robclark View Post
    By that reasoning, why bother working on opensrc drivers other than for nv or radeon hw, since those gpu's are faster still ;-)
    Good question indeed.


    • #12
      Originally posted by atom01 View Post
      Good question indeed.
      I'm not entirely sure if you completely missed my point, or if you are just trolling.


      • #13
        Please don't mind. I assure you guys that the majority of us appreciate the hard work you guys are doing. I myself have several products with differing versions of Adreno, so I am very much looking forward to that driver.


        • #14
          Originally posted by libv View Post
          What makes you believe that we are not working on the most friendly GPUs?

          I picked the Mali early 2011 as i assumed this had the most promising future ahead of itself, being ARMs own GPU. Rob Clark, working for Ti and Linaro at the time, chose Adreno, as his legal position towards Mali was questionable (linaro being largely sponsored by ARM), and he, like me for my N9 work, had access to the PowerVR source and then some. Both Erik Faye-Lund and Wladimir J Van Der Laan first approached me with the intent to work on PowerVR. I successfully convinced them to look elsewhere. Erik was first, and the Tegra was the best candidate then. When Wladimir came round, Vivante was simply waiting for him.

          Yes, we might seem to be spreading resources around, but at the same time we are limiting ourselves to the 4 sane GPUs, and ignoring PowerVR and other implementations. What happened for Videocore happened outside my sphere of influence, i personally would never have touched that hw with a 10ft pole, and i must applaud the fearless Videocore guys for undertaking a job which is even an order of magnitude harder than what we have to deal with on the 4 normal GPUs.

          Why 4 and not 1? Well, each of the people tackling each of those GPUs wanted to take this insane task from start to finish, not just finish what someone else already did. Plus, these 4 GPUs all have decent positions in the market, and they are sane and nice and each worthy of an open source driver. By spreading around across just those 4, we have a good chance of some degree of success on some. The way things are looking now means that we will be successful on all 4, but that we just need a bit of time to finish the massive task that is REing a GPU, and then writing a proper driver for it.

          So don't blame us for being slow on delivering, we are tackling an unbelievably time-consuming task, in our spare time. Don't be impatient if all you intend to do is talk about us not having delivered yet.
          Please don't take my comments as criticism. I'm in no position to tell you guys what to do and I certainly apreciate your work. My point was just trying to discuss the politics behind it all. You do have very valid points.


          • #15
            Originally posted by GreatEmerald View Post
            And Lima, well, it was first, I believe. And it got lucky with the whole optimised hardware deal.
            Well, I think, based on what libv has said and my experience (working on shaders), that we got lucky with the command stream and unlucky with the shaders. As libv proved with the Q3A demo, it's (relatively) easy to write an efficient driver for the mali cmdstream format. What flatmush and I discovered, though, is that the vertex shader and fragment shaders ISA's are much more complex and difficult to write an optimizing shader compiler for. I'm working on it though, and at the very least it won't be that hard to take what we have and write a very basic compiler for a Gallium driver.


            • #16
              You guys are all doing terrific work and making the GPU SoC world a good one. I'd gladly all buy you a beer next time I'm at fosdem


              • #17

                Interesting that the most popular embedded GPU (PowerVR) is also the one with the least amount of open-source work done on it. I do wonder why that is... Fear of being sued by Imagination Technologies? The people who are most qualified to do embedded GPU work all having worked with PowerVR chips and are covered by NDAs? Something about the way the PowerVR hardware works? Lack of people interested in "scratching the itch" and working on it? (If I had any skills in GPU driver development I would definatly be interested in contributing to a 100% open driver for the PowerVR GPU in my Nokia N900 but unfortunately I lack the skills)