3 Replies Latest reply on Jan 10, 2015 5:45 AM by boris

    CodeXL-bin linux user-space high cpu usage

    boris

      I am using linux mint with the amd drivers.

       

      Ahen i open codexl as a normal user it uses 100% of 1 core even when idling.

      But when i open it as a super user (root) it doesn't use any cpu performance.

       

      since i already had codeXL to profile, i profiled the user space codexl-bin program and the problem gets caused by

      Function, Samples, % of Hotspot Samples, Module

       

      FunctionSamples, % of Hotspot Samples,Module
      uki_firegl_MicroSleep836, 79.6949%libatiuki.so.1.0

       

      any idea to solve this?

      Or should i post this into the driver forum?

       

      update: i to have the exact same problem when i launch qt creator. so it might be a problem with qt/qtquick.

        • Re: CodeXL-bin linux user-space high cpu usage
          dorono

          Hi Boris,

          Can you share some details of your system?

          What CPU and GPU are you using? Which AMD Catalyst version (or other driver package) do you have installed? What Linux Mint version and what desktop manager?

          Can you run the profile you performed above with "Collect call stack details" enabled and share the callgraph of the uki_firegl_MicroSleep function?

           

          Thanks,

            • Re: Re: CodeXL-bin linux user-space high cpu usage
              boris
              Operating System Version

              Linux version 3.13.0-37-generic  linux mint 17.1

              windows managercinnemond

              amd driver

              app sdk

              CodeXL version

              latest 14.12

              Version, OpenCL 2.0 AMD-APP (1642.5)

              1.6.7247.0

              APU

              System Type, AMD A8-5550M APU with Radeon(tm) HD Graphics

              Renderer

              Version 4.4.13283 Compatibility Profile Context 14.501.1003
              dGPUdoesn't get detected properly but it states a "Hainan" GPU
              /home/boris/.amd/AMD/CodeXL/Project3_ProfilerOutput/Jan-04-2015_17-48-50

               

              i have attatched the profiling data, since you know the software better.

              the 18-00 was done with the assess profiling

              the 17-48 is an time sampling profiling

               

              the callstack points to

              Function (101 functions), Self Samples, Deep Samples, % of Deep Samples, No. of Paths, Source File, Module

              QMetaObject::cast(QObject*) const,           53,           54,      38.5714285714286, 33,                          , libQt5Core.so.5.3.0

               

              Can CodeXL be used with to profile an aplication using the library vexcl  since it generates the kernels on the fly and CloodeXL doesn't seem to be able to do a time trace profiling.

              • Re: CodeXL-bin linux user-space high cpu usage
                boris

                In the meantime i installed CodeXL onto my desktop machine, i used the exact same installation method and it works without a problem in userspace. However the desktop has an FX81xx chip with the R9 270x.

                 

                but i still wonder how i can profile a VexCL and Boost.compute based program.