Announcement

Collapse
No announcement yet.

Mesa 20.0's LLVMpipe Now Supports Running OpenCL On The CPU

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

  • Mesa 20.0's LLVMpipe Now Supports Running OpenCL On The CPU

    Phoronix: Mesa 20.0's LLVMpipe Now Supports Running OpenCL On The CPU

    Mesa's LLVMpipe Gallium3D driver has long been about running OpenGL on GPUs as a software fallback / debug path but as of this morning in Mesa 20.0-devel there is now the experimental ability of having OpenCL support making use of OpenCL "Clover" with NIR for CPU-based execution...

    Phoronix, Linux Hardware Reviews, Linux hardware benchmarks, Linux server benchmarks, Linux benchmarking, Desktop Linux, Linux performance, Open Source graphics, Linux How To, Ubuntu benchmarks, Ubuntu hardware, Phoronix Test Suite

  • #2
    Is there something like Mesa-Matrix for OpenCL? I would be quite interested in the current state of open source OpenCL.
    Honestly, I thought clover is dead ATM

    Comment


    • #3
      I don't mean to sound ungrateful about this, but, any chance that there will also be a CUDA equivalent? I personally can't think of any programs that depend on OpenCL, but there are some that do depend on CUDA, at least for feature-completion (off the top of my head, Meshroom is one of them). Not everyone has access to CUDA.
      Last edited by schmidtbag; 27 December 2019, 09:47 AM.

      Comment


      • #4
        Originally posted by schmidtbag View Post
        I don't mean to sound ungrateful about this, but, any chance that there will also be a CUDA equivalent? I personally can't think of any programs that depend on OpenCL, but there are some that do depend on CUDA, at least for feature-completion (off the top of my head, Meshroom is one of them). Not everyone has access to CUDA.
        OpenCL reminds me of the late OpenGL days when it kept pushing new releases while everyone was hoping for a reboot (which is now known as Vulkan) except that there's no plans to make a new API that could compete with CUDA.

        Comment


        • #5
          How does this work on drivers that do not support OpenCL? Could you use your standard graphics driver for 3D acceleration but fallback to the CPU for OpenCL emulation?

          Comment


          • #6
            Originally posted by Mani View Post
            Is there something like Mesa-Matrix for OpenCL? I would be quite interested in the current state of open source OpenCL.
            Honestly, I thought clover is dead ATM
            It's only "dead" in the sense that AMD stopped contributing to it.

            Comment


            • #7
              Is there a typo in the first sentence? Should it say "...running OpenGL on CPUs as a software fallback..." ?

              Comment


              • #8
                Originally posted by Mani View Post
                Is there something like Mesa-Matrix for OpenCL? I would be quite interested in the current state of open source OpenCL.
                Honestly, I thought clover is dead ATM
                Might want to take a look at Intel's Level 0 API that is part of their oneAPI initiative. It's like a lowerlevel OpenCL.

                This llvmpipe work is mainly so I can get some CI going around clover/spirv/nir for compute work, the mesa CL stack using SPIR-V isn't really useful yet, but there are a few more pieces of the puzzle that we have to drop in before it can do CL properly.

                Michael not sure it's worth testing before the other pieces of the puzzle are in place.

                Comment


                • #9
                  Well, this is all fine and good, but can you smoke the good stuff in the pipe?

                  That's all us old hippy engineers care about now

                  Comment


                  • #10
                    Originally posted by cl333r View Post
                    OpenCL reminds me of the late OpenGL days when it kept pushing new releases while everyone was hoping for a reboot (which is now known as Vulkan) except that there's no plans to make a new API that could compete with CUDA.
                    Vulkan can also be used for computing, so it can be seen as a successor to not only OpenGL but also OpenCL.

                    Comment

                    Working...
                    X