Skip to content
This repository has been archived by the owner on Feb 23, 2021. It is now read-only.

Latest commit

 

History

History
123 lines (102 loc) · 7.76 KB

BUILD.md

File metadata and controls

123 lines (102 loc) · 7.76 KB
Raw

CodeXL Build Instructions

Windows

Prerequisites

CodeXL requires JDK for CPU Profiling support of Java applications and the build may fail on Linux machines if JDK is not found in the system.

LPGPU2 CodeXL uses ADB to communicate with Android devices in order to start or stop the Remote Agent that runs on the application. ADB is not strictly required but it greatly simplifies the profiling experience.

Building CodeXL on Windows

  • Download and install the required software as listed in Prerequisites.
  • Clone the LPGPU2 CodeXL repository.
    • git clone https://github.com/codeplaysoftware/LPGPU2-CodeXL.git
  • Go to the folder Common/Lib/Ext/ComputeCpp and execute the script SetupComputeCpp.bat.
    • This will generate the file ComputeCpp-Version.props, which is required by the project LPGPU2SyclInfo.
  • Optional: Enabling build of CCVGCodeCoverage
    • The build of the CCVGCodeCoverage execution mode is disabled by default. In order to enabled it, perform the following steps:
      • Go to the folder CodeXL\Components\LPGPU2_CCVG\CCVGCodeCoveragePlugin
      • Double click on the script Enable_CCVGBuild.bat
  • Open the solution with Visual Studio 2015: LPGPU2-CodeXL\CodeXL\AllProjectsBuild\CodeXL-AllProjects.sln
    • In the Solution Explorer, right-click on the CodeXL-AllProjects solution and select Properties.
      • Open Common Properties > Startup Project.
      • Select the Single startup project option and set the project to AMDCodeXL.
      • Click OK.
  • Build 32 bit configuration in Release mode.
    • You must build CodeXL for the Win32 platform before you can build it for the x64 platform.
      • The Win32 build creates CodeXL(-d).lib, which is required for both Win32 and x64 CodeXL to run.
  • Build 64 bit configuration
  • If nothing goes wrong, CodeXL binaries will be located in LPGPU2-CodeXL\CodeXL\Output\Release\bin.
Installing CodeXL Drivers

After a successful build of CodeXL solution, you can install the drivers that CodeXL uses to power profile PC applications.

This step is optional for profiling LPGPU2 Android applications.

  • Go to the folder CodeXL\Components\CodeXLDrivers\Windows\Scripts
  • Run the script CodeXLInstallDriver_Standalone.bat

Linux

Preface
  • CodeXL is available in 32bit and 64bit flavours. We only support 64bit.
  • These build steps were performed on clean/new Ubuntu 16.04 64bit LTS machine.
  • If you are using a virtual Linux machine ensure the disk image size is at least 48GB.
  • CodeXL uses the SCons build system on Linux.
  • CodeXL require GCC version 4.8.5 or higher.
Prerequisites

We only support Ubuntu 16.04 64bit

  • Make sure the following packages are installed:
sudo apt install python2.7
sudo apt install adb
sudo apt install opencl-headers
sudo apt install gcc-multilib g++-multilib
sudo apt install libglu1-mesa-dev mesa-common-dev libgtk2.0-dev
sudo apt install zlib1g-dev libx11-dev
sudo apt install scons
sudo apt install libjpeg9-dev
sudo apt install libfltk1.3-dev
sudo apt install openjdk-8-jdk-headless
  • Make sure the environment variable JAVA_HOME is defined and pointing a to a valid JDK installation. If JAVA_HOME is not defined, the build will fail.
Building the experimental ComputeCpp Profiler
  • Download ComputeCpp from Codeplay's developer website
  • Extract the content in the folder LPGPU2-CodeXL/Common/Lib/Ext/ComputeCpp.
  • SCons will take care of finding ComputeCpp as long as it is extracted in the specified location.
Building CodeXL on Linux
  • cd LPGPU2-CodeXL/CodeXL/Util/linux
  • ./buildCodeXLFullLinuxProjects CXL_build=release LPGPU2_BUILD_CCVG=True -j8
  • If nothing goes wrong, CodeXL binaries will be located in LPGPU2-CodeXL/Output_x86_64/Debug/bin
Optional build of the Desktop GPU Backends:

In order to use the prototype of ComputeCpp SYCL Profiler on Linux, this step must be performed:

  • cd LPGPU2-CodeXL/Components/GpuProfiling/Build
  • ./backend_build.sh skip-32bitbuild skip-hsaprofiler
Build Switches
  • all SCons general switches, like -c for clean , more info at http://scons.org/doc/HTML/scons-man.html
  • -j specify the number of concurrent jobs (-j6).
  • LPGPU2_BUILD_CCVG=[True|False] - Enables or disables the build of the LPGPU2_CCVG project. By default the build is disabled.
  • CXL_build=[debug|release] - build type. If not stated default value is release.
  • CXL_build_verbose=1 - verbose output
  • CXL_boost_lib_dir=[path to boost libraries] - override the bundled boost libraries with files in given path
  • CXL_boost_include_dir=[path to boost headers] - override the bundled boost headers with files in given path
  • CXL_hsa=[true|false] - define if to build HSA parts. If not stated default value is false (skip HSA)
  • -c - performs a "clean" of all build targets.
  • When executing the backend_build.sh script, the following switches are supported:
    • debug: performs a debug build
    • skip-32bitbuild: skips building the 32-bit binaries
    • skip-oclprofiler: skips building the OpenCL profiler binaries
    • skip-hsaprofiler: skips building the HSA profiler binaries. If building on a system where you don't have HSA/ROCR header files, use this switch to avoid errors due to missing header files
    • hsadir 'dir': by default, when building the HSA Profiler binaries, the build scripts will look for the HSA/ROCR headers under /opt/rocm/hsa. You can override this location using the "hsadir" switch.
    • boostlibdir 'dir': by default, the build scripts will look for the boost libs in Common/Lib/Ext/Boost/boost_1_59_0/lib/RHEL6. You can override the location this location using the "boostlibdir" switch.
    • quick or incremental: performs an incremental build (as opposed to a from-scratch build)
    • clean: performs a "clean" of all build targets, removing all intermediate and final output files
  • LPGPU2_BUILD_ANDROID_PROTOCOL=[True|False] - Disable/Enable the building of the LPGPU2 remote device protocol and GUI. Default: True
Installing CodeXL Drivers

The power profiling driver is installed using the CodeXLPwrProfDriver.sh script, which is located at /CodeXL/Components/PowerProfiling/Backend/AMDTPowerProfilingDrivers/Linux. The script must be run with root privileges and accepts the following arguments:

  • install
  • uninstall

Note: When script will copy the CodeXLPwrProfDriverSource.tar.gz file to the /usr/src/amdtPwrProf-<version> directory, where it will be compiled.

Important

Before attempting to compile CodeXL if not on a clean machine be sure to clean out any previous builds of CodeXL especially if you have tried to build CodeXL with a compiler other than GCC. Calling buildCodeXLFullLinuxProjects -c to clean the build is not enough to clear out previously built libraries or object files. Likely the files will be used and not rebuilt. Manually remove the Output_x68_64 and recompile everything.