Announcement

Collapse
No announcement yet.

ROCm 1.8 Beta Packages Available For Radeon GPU Compute/OpenCL Testing

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

  • ROCm 1.8 Beta Packages Available For Radeon GPU Compute/OpenCL Testing

    Phoronix: ROCm 1.8 Beta Packages Available For Radeon GPU Compute/OpenCL Testing

    While ROCm 1.7.2 is the latest stable release for this Radeon GPU compute stack, there are 1.8.0 beta packages available for testing...

    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
    AMD, get OpenCL 2.0 (or better, 2.2) working. You have it with the fglrx driver, why not ROCm.

    Comment


    • #3
      Originally posted by marty1885 View Post
      AMD, get OpenCL 2.0 (or better, 2.2) working. You have it with the fglrx driver, why not ROCm.
      Using OpenCL 2.1 in 18.04 LTS (with flags --headless and --opencl=legacy for my RX 480 in the latest 18.20 Radeon Preview). Works on any kernel (4.15, 4.17rc etc)

      And the beauty of it is I'm still using Mesa stack with RADV and AMDGPU-PRO Vulkan alongside. Things are good right now.

      Code:
      $ clinfo | grep 'Platform Version\|Device Version\|Device Board Name\|Max compute units\|Max clock frequency\|Global memory size'
        Platform Version:          OpenCL 2.1 AMD-APP (2633.3)
        Max compute units:         36
        Max clock frequency:       1303Mhz
        Global memory size:        5439324160

      Comment


      • #4
        On my AMD platinum sponsored openSUSE Tumbleweed
        ( Platinum sponsors of openSUSE

        AMD

        )


        is not available ROCm as [ackage, yet as I know, Is there any possibility to activate,please?
        Thank you for your answer,

        clinfo

        Number of platforms 2
        Platform Name Clover
        Platform Vendor Mesa
        Platform Version OpenCL 1.1 Mesa 18.0.1
        Platform Profile FULL_PROFILE
        Platform Extensions cl_khr_icd
        Platform Extensions function suffix MESA

        Platform Name Portable Computing Language
        Platform Vendor The pocl project
        Platform Version OpenCL 1.2 pocl 1.0, LLVM 5.0.1
        Platform Profile FULL_PROFILE
        Platform Extensions cl_khr_icd
        Platform Extensions function suffix POCL

        Platform Name Clover
        Number of devices 1
        Device Name Radeon RX 550 Series (POLARIS12 / DRM 3.23.0 / 4.16.7-2.g938738d-default, LLVM 6.0.0)
        Device Vendor AMD
        Device Vendor ID 0x1002
        Device Version OpenCL 1.1 Mesa 18.0.1
        Driver Version 18.0.1
        Device OpenCL C Version OpenCL C 1.1
        Device Type GPU
        Device Available Yes
        Device Profile FULL_PROFILE

        Platform Name Portable Computing Language
        Number of devices 1
        Device Name pthread-AMD Ryzen 5 1500X Quad-Core Processor
        Device Vendor AuthenticAMD
        Device Vendor ID 0x1022
        Device Version OpenCL 1.2 pocl HSTR: pthread-x86_64-unknown-linux-gnu-bdver1
        Driver Version 1.0
        Device OpenCL C Version OpenCL C 1.2 pocl
        Device Type CPU
        Device Available Yes
        Device Profile FULL_PROFILE

        Comment


        • #5
          I believe the quick answer is "not yet" since our focus has been on pushing the required code upstream rather than building and testing packages for a broader range of distros.

          Longer answer is that most of the required KFD code is now upstream, although there is at least one more slug of commits in flight, and once that happens then it becomes a lot easier to start building ROCm stacks entirely from upstream source code - and at that point distros can start packaging the code and including it "off the shelf".
          Test signature

          Comment


          • #6
            Originally posted by Lucretia

            Is the LLVM/Clang code upstream yet?
            The llvm team works directly upstream. Patches generally flow upstream as soon as they are ready.

            Comment


            • #7
              I didn't think the atomic-less stuff involved llvm - is there some reason why you think that is the case ?
              Last edited by bridgman; 10 May 2018, 02:27 AM.
              Test signature

              Comment


              • #8
                I just received ROCm 1.8 on my Debian Sid machine so I guess it's out of beta already.
                Weirdly, it's not working though. I know I'm on an unsupported distro, but getting 1.7 working was relatively simple.
                Just remove 'firmware-amd-graphics' and compile, install and run the kernel from the ROCm GitHub repo instead of the vanilla Debian kernel and it would run fine.
                Not sure why it stopped working after the 1.8 update. Maybe I have to compile a new kernel? Doesn't look like anything changed in RadeonOpenCompute/ROCK-Kernel-Driver though.

                Comment


                • #9
                  My understanding was that the "atomic-less" stuff was primarily an MEC microcode change plus some changes to helper code in ROC runtime.

                  There would only be llvm involvement in atomics for the case where GPUs were submitting work to user queues themselves without going through CPU code, which AFAIK is supported but not used a lot yet.
                  Test signature

                  Comment

                  Working...
                  X