Announcement

Collapse
No announcement yet.

AMD To Exploit OpenCL Potential In LibreOffice

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

  • #16
    Originally posted by log0 View Post
    AMD OpenCL compiler seems to have problems with complex kernels. From what I've heard the compiler inlines all function calls and will blow up on such beasts as Cycles.
    What has AMD say about this?

    Comment


    • #17
      OpenCL support on r600 hw is making progress, even if the gallium compute page does not reflect it.
      Constant Address and local address are supported ; I'm not sure about private address, I think we replace private address with classic gpr/arrays. What's remaining is mostly synchronisation support and image read/write, although they require some work. Hopefully clover/r600g will support OpenCL 1.0 before the end of 2013.

      Comment


      • #18
        Originally posted by YAFU View Post
        Whenever you ask on blenderartists.org why AMD does not have good support for OpenCL on Cycles, they blame AMD and its uncooperative.
        Maybe AMD should try have a better relationship with the developers of Blender, as it is doing with LibreOffice developers.
        http://blenderartists.org/forum/show...AMD-and-Cycles
        They dont have an Problem with OpenCL because they use CUDA for Cycles and not OpenCL.

        Comment


        • #19
          Originally posted by Nille View Post
          They dont have an Problem with OpenCL because they use CUDA for Cycles and not OpenCL.
          That seems questionable for an open-source project...
          Is OpenCL really that worse then CUDA?
          (Maybe NVidia's implementation is... For a good reason.)

          I wouldn't be surprised if that CUDA dominance is mainly
          the result of good marketing by NVidia.

          Comment


          • #20
            Originally posted by blackiwid View Post
            I mean maybe from a developer persepective I could understand how somebody could want that, but from a user perspective that makes no sense... as user if you feel that way, just buy a nvidia card and let the amd guys there freedom.

            some users dont want nsa-backdoor-updates from nvidia...
            But Nvidia OpenCL performance suck, has they are based in CUDA, thus need to go through a compatibility layer.

            Anyway the OpenCL acceleration is mostly beeen developed for APU use, not dedicated GPU.

            Comment


            • #21
              Originally posted by Nille View Post
              They dont have an Problem with OpenCL because they use CUDA for Cycles and not OpenCL.
              The renderer supports both CUDA and OpenCL. I've seen videos of Cycles running on Nvidia gpus using OpenCL, even though the default is CUDA.

              Comment


              • #22
                Originally posted by Nille View Post
                What has AMD say about this?
                They are working on it, or something like that...

                But it seems to be a larger change affecting the whole compiler.

                Comment


                • #23
                  Originally posted by Nille View Post
                  They dont have an Problem with OpenCL because they use CUDA for Cycles and not OpenCL.
                  Originally posted by entropy View Post
                  That seems questionable for an open-source project...
                  No, you are wrong. Search the Internet and read about it, there's a lot written there.
                  Blender not choose CUDA over OpenCL. Just CUDA works, and there are problems with OpenCL. In fact, you can use OpenCL with Intel and nVidia in Cycles. With ATI too, but there are problems.
                  Blender needs AMD work with them to achieve a good performance in Cycles using OpenCL.

                  Comment


                  • #24


                    This picture sums it up with great eloquence.

                    You're using the fucking wrong tool for the job if your spreadsheet is that heavy.

                    Comment


                    • #25
                      Originally posted by curaga View Post
                      You're using the fucking wrong tool for the job if your spreadsheet is that heavy.
                      Lmao

                      To me it looks more like one of those "you've got this new shiny hammer and suddenly everything starts looking like a nail" problems.

                      Comment


                      • #26
                        linux needs something like grand central dispatch, where the OpenCL potential is used transparently and without much effort from the developer.

                        Comment


                        • #27
                          Originally posted by garegin View Post
                          linux needs something like grand central dispatch, where the OpenCL potential is used transparently and without much effort from the developer.
                          unless the wikipedia article on grand central dispatch is very wrong it has nothing to do with GPUs or OpenCL. Its just a nice library/language extension for writing multi-threaded code. Its more of a competitor to something like openMP.

                          Comment


                          • #28
                            That looks like a nice thing to work on. I know that one netbook I maintain could definitely use such a performance boost on LibreOffice.

                            Also, hey, this is specifically for HSA, so most likely Bridgman is working on it himself.

                            Originally posted by YAFU View Post
                            Are you kidding me, right?
                            blenderartists.org is much more active and with more users than any LibreOffice forum I know.
                            Forums are a very bad measure of activity. In order to do pretty much anything with Blender, you need to ask around on a forum. You don't need forums in order to achieve pretty much any tasks on LibreOffice.

                            Comment


                            • #29
                              Originally posted by GreatEmerald View Post
                              That looks like a nice thing to work on. I know that one netbook I maintain could definitely use such a performance boost on LibreOffice.

                              Also, hey, this is specifically for HSA, so most likely Bridgman is working on it himself.

                              Forums are a very bad measure of activity. In order to do pretty much anything with Blender, you need to ask around on a forum. You don't need forums in order to achieve pretty much any tasks on LibreOffice.
                              In support of that, I would suggest referring to Debian's popcon numbers:
                              http://qa.debian.org/popcon.php?package=blender
                              http://qa.debian.org/popcon.php?package=libreoffice

                              libreoffice-calc leads by far.

                              As far as Cycles OpenCL goes: see http://phoronix.com/forums/showthrea...883#post330883
                              Originally posted by Veerappan View Post
                              Cycles may be in the cards, but last time I tried to compile their shaders (several months ago), they were not working, and looked like they broke some of the CL standard... but it was hard to tell, given that it was all CUDA code that was ported/translated through pre-processor macros.

                              Comment


                              • #30
                                Originally posted by GreatEmerald View Post
                                Forums are a very bad measure of activity. In order to do pretty much anything with Blender, you need to ask around on a forum. You don't need forums in order to achieve pretty much any tasks on LibreOffice.
                                Well, for me the Blender forum is a good place for measurement. For example, you may know that there are much more than only 3 (ironic) users as "droidhacker" said (whom I had answered) buying the most expensive nVidia cards models on the market only for use with Cycles in Blender. In many cases, more than one card to build render farms. Also I can see in the forum that Blender is being increasingly used by many companies in the professional environment. So, to me that is a good way to measure something on the forum and is something that AMD/ATI should consider.

                                Comment

                                Working...
                                X