Piglit ------ 1. About 2. Setup 3. How to run tests 4. Available test sets 5. How to write tests 6. Todo 1. About -------- Piglit is a collection of automated tests for OpenGL and OpenCL implementations. The goal of Piglit is to help improve the quality of open source OpenGL and OpenCL drivers by providing developers with a simple means to perform regression tests. The original tests have been taken from - Glean ( http://glean.sf.net/ ) and - Mesa ( http://www.mesa3d.org/ ) 2. Setup -------- First of all, you need to make sure that the following are installed: - Python 2.7 or greater - Python Mako module - numpy (http://www.numpy.org) - cmake (http://www.cmake.org) - GL, glu and glut libraries and development packages (i.e. headers) - X11 libraries and development packages (i.e. headers) - waffle (http://people.freedesktop.org/~chadversary/waffle) Now configure the build system: $ ccmake . This will start cmake's configuration tool, just follow the onscreen instructions. The default settings should be fine, but I recommend you: - Press 'c' once (this will also check for dependencies) and then - Set "CMAKE_BUILD_TYPE" to "Debug" Now you can press 'c' again and then 'g' to generate the build system. Now build everything: $ make 2.1 Cross Compiling ------------------- On Linux, if cross-compiling a 32-bit build on a 64-bit host, then you must invoke cmake with option "-DCMAKE_SYSTEM_PROCESSOR=i386". 2.2 Ubuntu ---------- Install development packages. $ sudo apt-get install cmake g++ mesa-common-dev libgl1-mesa-dev python-numpy python-mako freeglut3-dev x11proto-gl-dev libxrender-dev Install additional components for which Ubuntu packages do not yet exist: - waffle (http://people.freedesktop.org/~chadversary/waffle) Configure and build. $ cmake . $ make 2.3 Mac OS X ------------ Install CMake. http://cmake.org/cmake/resources/software.html Download and install 'Mac OSX Universal' platform. Install Xcode. http://developer.apple.com/xcode Configure and build. $ cmake . $ make glean glean will not build with MacOSX10.7.sdk. If you are trying to build glean on Mac OS 10.7 (Lion), you will have to use MacOSX10.6.sdk. $ ccmake . Set 'CMAKE_OSX_SYSROOT' to '/Developer/SDKs/MacOSX10.6.sdk'. Configure. Generate and exit. $ make 2.4 Cygwin ---------- Install development packages. - cmake - gcc4 - make - opengl - libGL-devel - python - python-numpy - libglut-devel - libGLU-devel Configure and build. $ cmake . $ make 2.5 Windows ----------- Install Python. http://www.python.org/download Install NumPy. http://sourceforge.net/projects/numpy/files/NumPy Install CMake. http://cmake.org/cmake/resources/software.html Download and install 'Windows' platform. Install Microsoft Visual Studio. Install 'Visual C++' feature. Download OpenGL Core API and Extension Header Files. http://www.opengl.org/registry/#headers Copy header files to MSVC. C:\Program Files\Microsoft Visual Studio 10.0\VC\include\GL Download freeglut for MSVC. http://www.transmissionzero.co.uk/software/freeglut-devel Open Visual Studio Command Prompt. Start Menu->All Programs->Microsoft Visual Studio 2010->Visual Studio Tools->Visual Studio Command Prompt (2010) CD to piglit directory. Run CMake GUI. > C:\Program Files\CMake 2.8\bin\cmake-gui.exe . Configure - NMake Makefiles - Use default native compilers Set these variables in the Advanced view. - GLUT_INCLUDE_DIR - GLUT_glut_LIBRARY Configure Generate File->Exit Build from the Visual Studio Command Prompt. > nmake 3. How to run tests ------------------- Make sure that everything is set up correctly: $ ./piglit-run.py tests/sanity.tests results/sanity.results This will run some minimal tests. If you built Piglit out-of-source, then the environment variable PIGLIT_BUILD_DIR must be set: $ env PIGLIT_BUILD_DIR=/path/to/piglit/build/dir \ ./piglit-run.py tests/sanity.tests results/sanity.results Use $ ./piglit-run.py to learn more about the command's syntax. Have a look into the tests/ directory to see what test profiles are available: $ cd tests $ ls *.tests ... $ cd .. See also section 4. To create some nice formatted test summaries, run $ ./piglit-summary-html.py summary/sanity results/sanity.results Hint: You can combine multiple test results into a single summary. During development, you can use this to watch for regressions: $ ./piglit-summary-html.py summary/compare results/baseline.results results/current.results You can combine as many testruns as you want this way(in theory; the HTML layout becomes awkward when the number of testruns increases) Have a look at the results with a browser: $ xdg-open summary/sanity/index.html The summary shows the 'status' of a test: pass This test has completed successfully. warn The test completed successfully, but something unexpected happened. Look at the details for more information. fail The test failed. skip The test was skipped. [Note: Once performance tests are implemented, 'fail' will mean that the test rendered incorrectly or didn't complete, while 'warn' will indicate a performance regression] [Note: For performance tests, result and status will be different concepts. While status is always restricted to one of the four values above, the result can contain a performance number like frames per second] 4. Available test sets ---------------------- Test sets are specified as Python scripts in the tests directory. The following test sets are currently available: sanity.tests This suite contains minimal sanity tests. These tests must pass, otherwise the other tests will not generate reliable results. all.tests This suite contains all tests. all_cl.tests This suite contains all OpenCL tests. quick.tests Run all tests, but cut down significantly on their runtime (and thus on the number of problems they can find). In particular, this runs Glean with the --quick option, which reduces the number of visuals and state combinations tested. radeon.tests r300.tests r500.tests These test suites are adaptations of all.tests, with some tweaks to account for hardware limitations in Radeon chips. 5. How to write tests --------------------- Every test is run as a separate process. This minimizes the impact that severe bugs like memory corruption have on the testing process. Therefore, tests can be implemented in an arbitrary standalone language. I recommend C, C++ and Python, as these are the languages that are already used in Piglit. All new tests must be added to the all.tests profile. The test profiles are simply Python scripts. There are currently two supported test types: PlainExecTest This test starts a new process and watches the process output (stdout and stderr). Lines that start with "PIGLIT:" are collected and interpreted as a Python dictionary that contains test result details. GleanTest This is a test that is only used to integrate Glean tests Additional test types (e.g. for automatic image comparison) would have to be added to core.py. Rules of thumb: Test process that exit with a nonzero returncode are considered to have failed. Output on stderr causes a warning. 6. Todo ------- Get automated tests into widespread use ;) Automate and integrate tests and demos from Mesa Add code that automatically tests whether the test has rendered correctly Performance regression tests Ideally, this should be done by summarizing / comparing a history of test results Note that while some small artificial micro-benchmark could be added to Piglit, the Phoronix test suite is probably a better place for realistic performance testing.