Announcement

Collapse
No announcement yet.

AMD's Paying For Some Open-Source OpenCL Love

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

  • AMD's Paying For Some Open-Source OpenCL Love

    Phoronix: AMD's Paying For Some Open-Source OpenCL Love

    AMD's interested in driving the use of OpenCL within the open-source world, and they're willing to pay for it. One of their new contracts is to have more of the OpenCL work for GIMP/GEGL on the Open Computing Language...

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

  • #2
    Considering OpenCL will run on any graphics chip these days, I think this was pretty generous of AMD..

    Hopefully more people will see how much faster and more powerful doing these kinds of parallel floating point compute tasks on a 700 core/shader GPU is.

    Comment


    • #3
      when i read the title i thought that they were going to put money on Clover but anyway

      nice that they put resources in FOSS

      Comment


      • #4
        Originally posted by Sidicas View Post
        Considering OpenCL will run on any graphics chip these days, I think this was pretty generous of AMD..

        Hopefully more people will see how much faster and more powerful doing these kinds of parallel floating point compute tasks on a 700 core/shader GPU is.

        Very true, however, opencl is a rival of CUDA so even though opencl is nvidia compatible I'm sure stuff like this is to help fight against nvidia. Intel's graphics chips aren't really of concern. I'm sure this is also done because AMD needs more software publicity. Even though this may seem like a nice thing of AMD to do, Intel contributes to what seems to be dozens of open source projects, many of which are intended for Linux in mind.


        What I would find amusing is if Victor was using an intel and/or nvidia focused system, yet AMD pays him anyway.

        Comment


        • #5
          Originally posted by schmidtbag View Post
          Very true, however, opencl is a rival of CUDA so even though opencl is nvidia compatible I'm sure stuff like this is to help fight against nvidia. Intel's graphics chips aren't really of concern. I'm sure this is also done because AMD needs more software publicity. Even though this may seem like a nice thing of AMD to do, Intel contributes to what seems to be dozens of open source projects, many of which are intended for Linux in mind.


          What I would find amusing is if Victor was using an intel and/or nvidia focused system, yet AMD pays him anyway.
          It wouldn't be the first time... I was recently working on VP8+OpenCL acceleration for an arm-based mobile device, but all of my development work was being done on a Radeon 6850 (personal desktop) and GF 9400m (macbook). There are many parts of OpenCL programming that are entirely portable between architectures. It's just that you have to target specific vendor chips if you want to squeeze max performance out of the architectures. If you are just looking for faster-than-cpu performance, you can get by with generic GPU-based coding practices. If you need every last drop of performance, then you have to dig a bit deeper into the specific architecture you're optimizing for.

          Comment


          • #6
            Originally posted by Veerappan View Post
            It wouldn't be the first time... I was recently working on VP8+OpenCL acceleration for an arm-based mobile device, but all of my development work was being done on a Radeon 6850 (personal desktop) and GF 9400m (macbook). There are many parts of OpenCL programming that are entirely portable between architectures. It's just that you have to target specific vendor chips if you want to squeeze max performance out of the architectures. If you are just looking for faster-than-cpu performance, you can get by with generic GPU-based coding practices. If you need every last drop of performance, then you have to dig a bit deeper into the specific architecture you're optimizing for.
            What happened to this VP8+OpenCL acceleration of yours?

            Comment


            • #7
              Originally posted by Drago View Post
              What happened to this VP8+OpenCL acceleration of yours?
              http://github.com/awatry/libvpx.opencl

              I made a bunch of progress over the last few months (6000x improvement to loop filter speed), but I'm still not to the point where it's faster than CPU-only (currently 1/3 - 1/2 the speed of the hand-optimized assembly on my Phenom II x6 1055t + Radeon 6850 combination.... the CL path is close to the non-assembly generic C speed on the same hardware). Currently, 1080p videos are the only ones that give my desktop trouble. 720p and lower play at 100+ fps, but 1080p videos are in the 30-40 fps range.

              Currently, the parts that have OpenCL code paths are:
              - Subpixel Prediction (sixtap/bilinear filtering)
              - IDCT/Dequantization
              - Loop Filtering (de-blocking filter)

              The only one that is enabled by default is the loop filter, as I haven't had a chance to really optimize the other bits.

              As long as your machine has an OpenCL run-time library and headers installed, you should be able to build it.

              I've tested it on:
              AMD APP SDK (CPU and GPU)
              NVidia GPU SDK
              SNU OpenCL runtime (LLVM-based, GPL, http://opencl.snu.ac.kr/)

              I believe that you should be able to use this code on any device/platform that supports OpenCL 1.0 or higher. I don't have an Intel CPU, so I haven't been able to test their run-time, and I haven't yet gotten Denis' clover branch to work yet (not that his code is broken, I just haven't gotten the LLVM build working yet).

              I'll probably push the current git repository to my WebM sandbox in the future, and hopefully I can get this code accepted upstream after the encoder part of the library has been tested for regressions/breakages.

              --Aaron

              Comment


              • #8
                Hi! I'm Victor, maybe the title of the article is a little misleading. I think open-source community is very excited about OpenCL in general and many projects already use it like Luxrender and Darktable, specially if we consider that there isn't yet an stable open-source implementation [but I suppose this will change soon as the work on Clover and others progress]. AMD support [technical and financial] is very helpful and is a sign of their good will to the community.

                Besides, I'm very excited with this project and I hope everything goes well and you like the results. Notice that through GEGL we are able to improve not just GIMP, but others projects like Darktable and Mypaint. Feel free to ask anything

                Victor Oliveira

                Comment


                • #9
                  Congratulations to Victor for such a great opportunity

                  and

                  a big thanks to AMD for their great open source initiative.

                  Comment


                  • #10
                    @victormoliveira

                    i haz question


                    why they decided to fund GEGL???. i mean its important for many projects but its not something really important like infrastructure (ie clover, openGL 4, g3d video etc)

                    Comment


                    • #11
                      They already have an infrastructure, it's called Catalyst. Yes it sucks. But I don't think this is about supporting open source or Linux, I think this is about promoting the use of OpenCL in apps, because only if apps start leveraging the GPU they can compete with Intel.

                      Comment


                      • #12
                        what short memory's you have

                        :wave:

                        Michael said:


                        "AMD's interested in driving the use of OpenCL within the open-source world, and they're willing to pay for it. One of their new contracts is to have more of the OpenCL work for GIMP/GEGL on the Open Computing Language. "

                        ..."GSoC has obviously been over for a few months, but there's now word that AMD is financing the student developer (Victor Oliveira) to keep working on this code. "I've been contacted by AMD and they are very interested in OpenCL adoption by the Open-Source community and want to help me to continue this project. So, I'll be continuing it as a contractor for AMD in the next few months. Of course, all code will be Open-Source. In fact, I've been developing it in my Gegl branch." ....
                        "

                        "AMD's interested in driving the use of OpenCL within the open-source world" indeed, did you forget already what
                        a mess last chrismas's SDK release turned into where they included the OVDecode.lib for the windows 32bit install, but OVDecode.so Never actually appeared in the 32 bit linux lib/x86 dir, did that change all of a sudden today as regards the OpenCL OpenVideo driver for video decode ( the only part many of us care about as a first step), a full year later !

                        http://phoronix.com/forums/showthrea...163#post161163
                        gbeauche said: "OpenVideo has public headers and private (or inexistent) libraries for Linux. Ever seen something right at ATI? "

                        etc....

                        never mind paying for it, how about they actually release compile the damned OVDecode code for linux use and give it some DCAA

                        Comment


                        • #13
                          Amd has got huge performance problems as the speed/core is not good. But compared to intel they have got the faster gpu. Amd officially dislikes cpu only benchmarks - they hate sysmark as their cpus win absolutely nothing against Intel. So they NEED software that can use the gpu. They do nothing else than Nvidia, they even helped to produce CUDA apps because those do not run on Amd gpus. In order to win some benchmarks AMD needs OpenCL, so they need an app that really speeds up using it. Paying a developer to create what they need is somehow logical, but the idea comes a bit late

                          Comment


                          • #14
                            Originally posted by Kano View Post
                            Amd has got huge performance problems as the speed/core is not good. But compared to intel they have got the faster gpu. Amd officially dislikes cpu only benchmarks - they hate sysmark as their cpus win absolutely nothing against Intel. So they NEED software that can use the gpu. They do nothing else than Nvidia, they even helped to produce CUDA apps because those do not run on Amd gpus. In order to win some benchmarks AMD needs OpenCL, so they need an app that really speeds up using it. Paying a developer to create what they need is somehow logical, but the idea comes a bit late
                            you are right its only logic and no its never to late.

                            Comment


                            • #15
                              Originally posted by Kano View Post
                              Amd has got huge performance problems as the speed/core is not good.
                              Yeah, provided you're using this logic:

                              take a single-threaded benchmarks where AMD ties Intel, normalize to 8 core vs 4 core(because hyperthreading is magically much less like a real core in this particular scenario than it normally is when Intel fans are bragging about it), then exclaim "well, it took AMD twice as many cores just to tie Intel, so they're half as fast per core".

                              Although Llano and Trinity should certainly stomp anything Intel has when properly utilizing GPU compute, but when AMD benchmarks 8 x faster than Intel in Gimp3, everyone will start counting shaders as cores; "OMG, AMD needed 120x more cores to beat Intel by a mere 800%!!!! FAIL!!!!!"

                              Comment

                              Working...
                              X