Announcement

Collapse
No announcement yet.

LibreOffice Lands A Ton Of GPU OpenCL Functions

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

  • #31
    Originally posted by ciplogic View Post
    It looks to me that your use case is document specific. At least for me no documents do load so slow. Did you create a bug in LO's bug database? As for accelerating using OpenCL I would say that gives another way to improve some uses of LO and I think that no one should complain. If there will be accelerated just for AMD systems, maybe will convince people with slow LO to use it later if they have an APU.

    Look to this blog post, maybe fixes your issue http://www.lanedo.com/2013/how-your-...e-contributor/
    Besides the financial math/linear programming, engineering mathematics,etc the Graphics back-end can most certainly benefit from OpenCL acceleration as more of it goes OpenGL based. And no, OpenMP won't help in this regard.

    Comment


    • #32
      Originally posted by Marc Driftmeyer View Post
      Besides the financial math/linear programming, engineering mathematics,etc the Graphics back-end can most certainly benefit from OpenCL acceleration as more of it goes OpenGL based. And no, OpenMP won't help in this regard.
      openCL yes, but no using spreadsheets LOL. And openMP does helps in those areas too!

      Comment


      • #33
        Originally posted by Alex Sarmiento View Post
        openCL yes, but no using spreadsheets LOL. And openMP does helps in those areas too!
        I hve to say that OpenMP helps much less than OpenCL. Because OpenMP requires to write software with no dependencies so it require a refactor in this regard and needs to be implemented by someone interested. OpenCL interests the OpenCL vendors (here AMD) and OpenCL gives "OpenMP for free" in the sense that OpenCL code can be computed on many CPU cores. Even OpenCL is more limited than OpenMP, in fact what is ported to OpenCL has support for multicore. If you or anyone from Intel will rewrite parts to take advantage of multicore (using OpenMP or C++ 14's async), for weak cores, fast GPU will be little speedup, but the OpenCL codes will matter more (at least for that particular machine).

        Comment


        • #34
          Originally posted by mmstick View Post
          I find it funny that people fail to realize the implications of this. This means even lower end devices can use spreadsheets, and create them larger. It means spreadsheets load quicker and expend less energy because OpenCL calculations on a GPU is more energy efficient than carrying it out on a CPU. If you have an AMD APU system, more power. If your Android device supports OpenCL with it's GPU, it's a no brainer.
          No, it's far from that clear cut, not even on APUs.

          There is a cost to transferring things to GPU memory. There is a cost to controlling that GPU, and a cost to powering the GPU to higher power states. So for small jobs you end up with a loss, both in power used and in time taken.

          This means even lower end devices can use spreadsheets
          Also, low-end devices do not support OpenCL at all...

          Comment


          • #35
            Originally posted by schmidtbag View Post
            For the record, when I'm complaining about Calc's slowness, there have been times I waited over a minute for something to finish. On Excel, only a few seconds. I think Calc is a great piece of software, I just think OpenCL will be a major benefit to it.
            Please try LO Calc 4.1.2 it's much much faster than LO Calc's 3.6 and 4.0!
            Last edited by ahmad; 10-31-2013, 05:31 AM. Reason: added calc

            Comment


            • #36
              Originally posted by curaga View Post
              No, it's far from that clear cut, not even on APUs.

              There is a cost to transferring things to GPU memory. There is a cost to controlling that GPU, and a cost to powering the GPU to higher power states. So for small jobs you end up with a loss, both in power used and in time taken.


              Also, low-end devices do not support OpenCL at all...
              They might in the future, though. Notice how both AMD and intel are integrating increasingly useful 3D cores on their CPUs, and how these integrated GPUs have rather fast interconnects and a (so far fairly small) chunk of on-die RAM to speed things up. Both AMD and intel have been vaguely hinting that integrated GPUs will take over some of the role that SSE and similar extensions have fulfilled so far, and if they actually want that to happen it would make sense to have openCL support even on their ultra-low-power chips.

              That said, I have no idea what the state of things are on ARM. Given how bad the graphics drivers apparently are, I doubt we'll get commonplace and good openCL support on ARM anytime soon.

              Comment


              • #37
                Originally posted by oleid View Post
                Actually, that is what I think, too. Proof of concept for AMD. Nevertheless, this code might come handy for some people.
                Actually, MS-Office 2010 implements the same thing, so I'd rather think they are working on feature parity, not a proof of concept.

                Comment


                • #38
                  Originally posted by mmstick View Post
                  This means even lower end devices can use spreadsheets, and create them larger.
                  I remember using spreadsheets on 40MHz SPARC that was emulating a 286 running Windows. I don't think you could find anything much lower-end than that these days.

                  I can't help but feel that if you need GPU acceleration for your spreadsheets on a multi-core, multi-gigahertz CPU, you're probably doing it wrong.

                  Comment


                  • #39
                    Originally posted by movieman View Post
                    I remember using spreadsheets on 40MHz SPARC that was emulating a 286 running Windows. I don't think you could find anything much lower-end than that these days.

                    I can't help but feel that if you need GPU acceleration for your spreadsheets on a multi-core, multi-gigahertz CPU, you're probably doing it wrong.
                    The spreadsheets companies use are AWFUL, AWFUL, AWFUL. But you know what? They work. And trying to convince your boss to spend $10,000 to rewrite one as an actual app with a real DB is an exercise in wasting your time, because the people who use the spreadsheets are the ones who feel completely comfortable working with a spreadsheet but don't know which end of a mouse is supposed to point up. Because the spreadsheet is all they've ever known, and they're just monkeys repeating what they've been trained to do without any actual thought process.

                    So yes, this type of advance is very useful, to a lot more people than you guys would ever expect. And yes, it sucks that this is true, because they ARE doing it wrong. They just don't care.

                    It's mostly useless for home users, because they don't use spreadsheets that way.
                    Last edited by smitty3268; 11-02-2013, 02:56 AM.

                    Comment


                    • #40
                      Originally posted by smitty3268 View Post
                      Because the spreadsheet is all they've ever known, and they're just monkeys repeating what they've been trained to do without any actual thought process.
                      A lot of the people using the spreadsheets aren't programmers - they've got a system that works and that they know how to use. Convincing anybody to change from that state of affairs (or telling them that they need to learn a new system in order to do what they've been doing for years) takes a lot of effort, and can cause resentment, particularly if they can't see a benefit for themselves - see Microsoft's ribbon interface in Office 2007.

                      Comment

                      Working...
                      X