Announcement

Collapse
No announcement yet.

Benchmarking Radeon Open Compute ROCm 1.4 OpenCL

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

  • Benchmarking Radeon Open Compute ROCm 1.4 OpenCL

    Phoronix: Benchmarking Radeon Open Compute ROCm 1.4 OpenCL

    Last month with AMD/GPUOpen's ROCm 1.4 release they delivered on OpenCL support, albeit for this initial release all of the code is not yet open-source. I tried out ROCm 1.4 with the currently supported GPUs to see how the OpenCL performance compares to just using the AMDGPU-PRO OpenCL implementation.

    http://www.phoronix.com/vr.php?view=24028

  • #2
    Nice testing!

    I really don't understand all the components though...
    KFD
    HSA
    HSA+
    Rock kernel driver,
    the other KFD that "is NOT compatible with amdkfd that is distributed as part of the mainline Linux kernel from 3.19 and onward"
    kmt,
    rocr runtime,
    hsa runtime,
    hcc,
    thunk.

    Comment


    • #3
      What about OpenMP?
      https://www.phoronix.com/forums/foru...808#post921808

      In pair with Down-CC this particular approach looks very promising.

      Comment


      • #4
        Originally posted by phoronix View Post
        Phoronix: Benchmarking Radeon Open Compute ROCm 1.4 OpenCL

        Last month with AMD/GPUOpen's ROCm 1.4 release they delivered on OpenCL support, albeit for this initial release all of the code is not yet open-source. I tried out ROCm 1.4 with the currently supported GPUs to see how the OpenCL performance compares to just using the AMDGPU-PRO OpenCL implementation.

        http://www.phoronix.com/vr.php?view=24028
        Can the AMDGPU-PRO kernel module and the ROCm kernel module be used on a single Linux machine at the same time?

        Is ROCm compatible with the open-source amdgpu.ko in the Linux kernel tree?

        Comment


        • #5
          Originally posted by atomsymbol View Post

          Can the AMDGPU-PRO kernel module and the ROCm kernel module be used on a single Linux machine at the same time?
          Didn't try that.
          Originally posted by atomsymbol View Post
          Is ROCm compatible with the open-source amdgpu.ko in the Linux kernel tree?
          I don't think so, at least not yet. Just like I don't believe it's AMDKFD kernel module is compatible with the mainline one right now, hence it ships its own Linux 4.6 kernel as part of its Ubuntu 16.04 packages.
          Michael Larabel
          http://www.michaellarabel.com/

          Comment


          • #6
            Originally posted by ernstp View Post
            Nice testing!

            I really don't understand all the components though...
            KFD
            HSA
            HSA+
            Rock kernel driver,
            the other KFD that "is NOT compatible with amdkfd that is distributed as part of the mainline Linux kernel from 3.19 and onward"
            kmt,
            rocr runtime,
            hsa runtime,
            hcc,
            thunk.
            I believe very few people other than the developers understand those.

            Comment


            • #7
              I thought rocm used exactly the same opencl components as the AMDGPU-pro driver. I remembered Bridgman saying something along those lines, but I probably miss read his comment then

              Comment


              • #8
                Source code or it didn't happen.
                ## VGA ##
                AMD: X1950XTX, HD3870, HD5870
                Intel: GMA45, HD3000 (Core i5 2500K)

                Comment


                • #9
                  Installed on Debian resulting in segfaults for CodeXL that needs an update to reflect ROCm. Uninstalled and CodeXL works once more.

                  Comment


                  • #10
                    Originally posted by chimpy View Post
                    I thought rocm used exactly the same opencl components as the AMDGPU-pro driver. I remembered Bridgman saying something along those lines, but I probably miss read his comment then
                    It's LLVM is LLVM-amdgpu 3.9 from AMD, so there is definitely customization going on with all the APIs and HSA subcomponents.

                    Comment

                    Working...
                    X