Skip to content

Graph algorithms for machine learning frameworks

License

Notifications You must be signed in to change notification settings

graphcore/poprithms

Repository files navigation

Overview

Poprithms is a graph algorithms library used by ML frameworks. Algorithm specific information can be found in the notes sub-directory.

Prerequisites for building

  • Boost. Tested with versions 1.70 and 1.78, but will probably work with newer and older versions.

Prerequisites for building with optional extensions

  • GoogleTest. Some tests and mock classes use GoogleTest. By default, these are built. To disable, toggle the cmake flag POPRITHMS_BUILD_MOCKS.

Configure with cmake

Create a build directory:

mkdir build; cd build;

Configure cmake. If boost is installed in a standard location, from the build directory,

cmake /path/to/poprithms/root/dir

If boost is installed somewhere unusual,

cmake /path/to/poprithms/root/dir -DBOOST_ROOT=/path/to/boost/install 

By default, the C++ compiler flag -Werror is enabled. This escalates warnings to errors. To disable -Werror, use

cmake /path/to/poprithms/root/dir  -DPOPRITHMS_WERROR=OFF

By default, Boost Stacktrace is used to improve error messages. To disable it,

cmake /path/to/poprithms/root/dir  -DPOPRITHMS_USE_STACKTRACE=OFF

By default, building the google mock library for poprithms is enabled. When enabled, cmake will try to find the GTest package to use in the poprithm google mocks. To disable it,

cmake /path/to/poprithms/root/dir -DPOPRITHMS_BUILD_MOCKS=OFF

The usual CMake flags can be used to set the poprithms install directory CMAKE_INSTALL_PREFIX , and the generator CMAKE_GENERATOR. A complete configuration command might be:

cmake /path/to/poprithms/root/dir  -DPOPRITHMS_WERROR=OFF -BOOST_ROOT=/path/to/boost/install -DCMAKE_INSTALL_PREFIX=/my/install/dir -DCMAKE_GENERATOR="Ninja"

Build the library

Once configured, the library can be built from the build directory. If the generator is Ninja and you have an install directory set, then

ninja install

will build and copy the poprithms shared library (no static library is built), header, and configuration files into the install directory.

Build the documentation

Currently poprithms does not build documentation.

Examples

Currently the test directory serves as examples.

Using the poprithms library in CMake projects

The easiest way to use poprithms in CMake is via find_package, specifying on the CMAKE_PREFIX_PATH the location of the poprithms install.

find_package(poprithms CONFIG REQUIRED)
target_link_libraries(my_library PRIVATE poprithms)

Formatting the C++ code

The project currently uses clang-format version 14.

Troubleshooting

CMake can't find poprithms

An error like the following is caused when poprithms-config.cmake is not on the defined set of search paths.

CMake Error at CMakeLists.txt:X (find_package):
  By not providing "Findpoprithms.cmake" in CMAKE_MODULE_PATH this project
  has asked CMake to find a package configuration file provided by
  "poprithms", but CMake did not find one.

  Could not find a package configuration file provided by "poprithms" with
  any of the following names:

    poprithmsConfig.cmake
    poprithms-config.cmake

  Add the installation prefix of "poprithms" to CMAKE_PREFIX_PATH or set
  "poprithms_DIR" to a directory containing one of the above files.  If
  "poprithms" provides a separate development package or SDK, be sure it has
  been installed

To fix this, specify poprithms_DIR (or poprithms-static_DIR for the static library) as a environment or CMake variable when configuring CMake to the location of poprithms-config.cmake.

$ poprithms_DIR=XXX/install/poprithms/lib/cmake/ cmake ..

where XXX is the path to the poprithms install.

See https://cmake.org/cmake/help/latest/command/find_package.html for more information.

About

Graph algorithms for machine learning frameworks

Topics

Resources

License

Stars

Watchers

Forks

Packages

No packages published

Languages