Announcement

Collapse
No announcement yet.

Radeon Open Compute 1.3 Platform Brings Polaris & Other Features

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

  • #61
    Good morning!
    I am a little confused not finding kaveri in the list of supported platforms. Is it still being supported? If so does clblas (or rocblas) work with it?

    Comment


    • #62
      Originally posted by Amarildo View Post
      *PLEASE EXPLAIN HOW* hehehehehehehehhehe
      Originally posted by mibo View Post
      I also would like to know how to set up this?
      This is how I did it. It says opensuse but it's working for me on gentoo. I think it works with any version of the opencl driver.

      Warning: I am not responsible for you bricking your system.


      Mystro256 also has an example rpm to look at:
      GitHub is where people build software. More than 100 million people use GitHub to discover, fork, and contribute to over 420 million projects.

      Comment


      • #63
        Originally posted by fuzz View Post



        This is how I did it. It says opensuse but it's working for me on gentoo. I think it works with any version of the opencl driver.

        Warning: I am not responsible for you bricking your system.


        Mystro256 also has an example rpm to look at:
        https://github.com/Mystro256/amdgpu-pro-opencl-rpm
        Dammit, only GCN 1.1 and higher!!!!

        Comment


        • #64
          IIRC at least one user has experimented with -pro OpenCL on amdgpu over SI. Maybe haagch ?
          Last edited by bridgman; 21 November 2016, 04:42 PM.
          Test signature

          Comment


          • #65
            Originally posted by fuzz View Post



            This is how I did it. It says opensuse but it's working for me on gentoo. I think it works with any version of the opencl driver.

            Warning: I am not responsible for you bricking your system.


            Mystro256 also has an example rpm to look at:
            https://github.com/Mystro256/amdgpu-pro-opencl-rpm
            Thanks a lot for the link. It is working perfectly on my Debian Testing/Experimental.

            Comment


            • #66
              Originally posted by bridgman View Post
              IIRC at least one user has experimented with -pro OpenCL on amdgpu over SI. Maybe haagch ?
              Thanks, but do you remember what the results were?

              Comment


              • #67
                For information, I'm using the OpenCL stack from amdgpu-pro since months using the trick above, and it's super stable and very-efficicient. In fact I haven't updated to the last amdgpu-pro release because I don't need it and the OpenCL-only installation is tricky, and since it just works, I don't need to update it. I can update my whole distro without having to care about that. I updated from Ubuntu 16.04 to 16.10 without breaking my OpenCL setup. I'm upgrading my Mesa stack (using Padoka PPA) every week without breaking my OpenCL setup. It works awesomely.

                My GPU is a CIK one, not a SI one, but as said in the topic by others, in the worst case SI owner will get the same as I have today. Even if SI owner will never get the ROC-based opensource OpenCL stack, SI owner will be able to get amdgpu-pro opencl-only which is very good. SI owner will get a nice, functional and performant OpenCL stack. The installation just have to be made easier, as bridgman said. The ones who fear about AMD proprietary OpenCL stuff because of AMD proprietary graphic stuff reputation don't have to fear. AMD proprietary graphic stack is pure crap, AMD proprietary compute stack is wery good, stable, performant and nice (the only stuff needed is openness). Don't judge the compute stack with graphic stack reputation. And the compute stack is usable without having to use the graphic stack.

                This is my graphic and compute stack:

                Code:
                Model:         AMD Hawaii XT / Grenada XT “Radeon R9 290X/390X”
                Shader:        1080.000 MHz (high/performance)
                Memory:        1500.000 MHz (8192M)
                KMD:           linux/amdgpu 4.8
                UMD:           mesa/amdgpu 13.1
                OpenCL:        1.2 (amdgpu-pro 2117.7))
                OpenGL:        4.5 (mesa/amdgpu 13.1)
                OpenGL ES:     3.1 (mesa/amdgpu 13.1)
                Vulkan:        1.0 (mesa/radv 13.1)
                I use OpenCL for darktable rendering and I'm very happy with that setup.

                Comment


                • #68
                  illwieckz I've downloaded the ready-to-use files, but Blender just crashed when I went to the "System" tab (where I specify the GPU for compute). I'm yet to run it with gdb.

                  Code:
                  # Blender 2.78 (sub 0), Commit date: 2016-10-24 12:20, Hash e8299c8
                  
                  # backtrace
                  blender(BLI_system_backtrace+0x30) [0x13ec140]
                  blender() [0xa21bb9]
                  /usr/lib/libc.so.6(+0x330b0) [0x7f6122b1a0b0]
                  /usr/local/lib64/libdrm_amdgpo.so.1(amdgpu_query_gpu_info+0x12) [0x7f60e664ad32]
                  /usr/local/lib64/libamdocl64.so(+0xa25fee) [0x7f60e98b4fee]
                  /usr/local/lib64/libamdocl64.so(+0xa2665b) [0x7f60e98b565b]
                  /usr/local/lib64/libamdocl64.so(+0xa28c2b) [0x7f60e98b7c2b]
                  /usr/local/lib64/libamdocl64.so(+0xa18648) [0x7f60e98a7648]
                  /usr/local/lib64/libamdocl64.so(+0x70ed53) [0x7f60e959dd53]
                  /usr/local/lib64/libamdocl64.so(+0x690df9) [0x7f60e951fdf9]
                  /usr/local/lib64/libamdocl64.so(+0x690e57) [0x7f60e951fe57]
                  /usr/local/lib64/libamdocl64.so(+0x691ba9) [0x7f60e9520ba9]
                  /usr/local/lib64/libamdocl64.so(+0x657f54) [0x7f60e94e6f54]
                  /usr/local/lib64/libamdocl64.so(+0x6593e7) [0x7f60e94e83e7]
                  /usr/local/lib64/libamdocl64.so(+0x659576) [0x7f60e94e8576]
                  /usr/local/lib64/libamdocl64.so(+0x618ca0) [0x7f60e94a7ca0]
                  /usr/local/lib64/libamdocl64.so(+0x632eb7) [0x7f60e94c1eb7]
                  /usr/local/lib64/libamdocl64.so(clIcdGetPlatformIDsKHR+0x73) [0x7f60e9491493]
                  /usr/lib/libOpenCL.so(+0x6019) [0x7f60ecb6d019]
                  /usr/lib/libOpenCL.so(clGetPlatformIDs+0x13c) [0x7f60ecb6dd7c]
                  blender() [0x1957762]
                  blender(_ZN3ccl18device_opencl_infoERNS_6vectorINS_10DeviceInfoENS_16GuardedAllocatorIS1_EEEE+0x53) [0x19588c3]
                  blender(_ZN3ccl6Device17available_devicesEv+0xf1) [0x1936ee1]
                  blender() [0x183d838]
                  blender() [0x1364791]
                  blender(RNA_property_enum_items_ex+0xe0) [0x1258a30]
                  blender(RNA_property_enum_items+0x13) [0x1258a53]
                  blender(RNA_property_enum_identifier+0x36) [0x125fc36]
                  blender() [0xe1dd3b]
                  blender(pyrna_prop_to_py+0x1c8) [0xe278a8]
                  blender() [0xe2797f]
                  /usr/lib/libpython3.5m.so.1.0(+0x11c6f7) [0x7f612d9336f7]
                  /usr/lib/libpython3.5m.so.1.0(PyCFunction_Call+0xc9) [0x7f612d8c6bc9]
                  /usr/lib/libpython3.5m.so.1.0(PyEval_EvalFrameEx+0x86fb) [0x7f612d93e79b]
                  /usr/lib/libpython3.5m.so.1.0(+0x1285e3) [0x7f612d93f5e3]
                  /usr/lib/libpython3.5m.so.1.0(PyEval_EvalCodeEx+0x23) [0x7f612d93f6c3]
                  /usr/lib/libpython3.5m.so.1.0(+0x92fd8) [0x7f612d8a9fd8]
                  /usr/lib/libpython3.5m.so.1.0(PyObject_Call+0x47) [0x7f612d87f2b7]
                  blender() [0xe26a8e]
                  blender() [0x13553ef]
                  blender(ED_region_panels+0x392) [0xd08412]
                  blender(ED_region_do_draw+0xa77) [0xd07407]
                  blender() [0xa25e64]
                  blender(wm_draw_update+0x5de) [0xa2715e]
                  blender(WM_main+0x28) [0xa22818]
                  blender(main+0x40f) [0xa029af]
                  /usr/lib/libc.so.6(__libc_start_main+0xf1) [0x7f6122b07291]
                  blender(_start+0x2a) [0xa1e8ba]

                  Comment


                  • #69
                    I've created a working dir called "/usr/local/lib64" (which didn't exist before), then:

                    Code:
                    [root@amarildo lib64]# ln -s libamdocl64.so libOpenCL.so
                    [root@amarildo lib64]# ln -s libamdocl64.so libOpenCL.so.1
                    [root@amarildo lib64]# ln -s libamdocl64.so libOpenCL.so.1.0.0
                    Code:
                    [root@amarildo lib64]# ls -l
                    total 96088
                    -rwxr--r-- 1 root root 37384760 Mar 17  2016 libamdocl12cl64.so
                    -rwxr--r-- 1 root root 60947576 Mar 22  2016 libamdocl64.so
                    lrwxrwxrwx 1 root root       22 Nov 23 20:54 libdrm_amdgpo.so.1 -> libdrm_amdgpo.so.1.0.0
                    -rwxr--r-- 1 root root    38144 Mar 17  2016 libdrm_amdgpo.so.1.0.0
                    lrwxrwxrwx 1 root root       14 Nov 23 20:56 libOpenCL.so -> libamdocl64.so
                    lrwxrwxrwx 1 root root       14 Nov 23 20:56 libOpenCL.so.1 -> libamdocl64.so
                    lrwxrwxrwx 1 root root       14 Nov 23 20:56 libOpenCL.so.1.0.0 -> libamdocl64.so
                    [root@amarildo lib64]#
                    I also installed 'ocl-icd' and created the following symlinks it, but it didn't work:

                    /usr/lib/libOpenCL.so -> /usr/local/lib64/libamdocl64.so
                    /usr/lib/libOpenCL.so.1 -> /usr/local/lib64/libamdocl64.so
                    /usr/lib/libOpenCL.so.1.0.0 ->/usr/local/lib64/libamdocl64.so

                    Comment


                    • #70
                      I'll try the same with Catalyst's OpenCL.

                      Comment

                      Working...
                      X