Building the Project

Requirements

To use this project CMake is required. You can check if CMake is installed by running

cmake --version

Note

CMake can be downloaded or installed OS specific through other methods

Linux

A simple way is using apt-get by running the command

sudo apt-get install cmake

Building

1. Download the git repository with git

   HTTPS

   git clone https://github.com/Integer-Ctrl/machine-learning-compilers
   

   SSH

   git clone git@github.com:Integer-Ctrl/machine-learning-compilers
   

2. Go into the project folder. Your current path should look like this ../machine-learning-compilers.

3. Now create a new folder called build with

   mkdir build
   

4. Go into this directory. Your current path should look like this ../machine-learning-compilers/build.

5. Now we can start with CMake. Run the following command

   cmake .. -DCMAKE_BUILD_TYPE=<Type>
   

   Supported values for <Type> are Release and Debug. If only cmake .. is used the Release build is selected.

   The most desired command might be:

   cmake .. -DCMAKE_BUILD_TYPE=Release
   

   Note

   With the Option -G a Generator can be defined used to create the make files and compile the Code. All available Generators can be found at the bottom of the
   cmake --help text. An Example could look like this

   cmake .. -G "MinGW Makefiles"
   

   Important

   When using a multi-config Generator, i.e. Ninja Multi-Config, Visual Studio Generators, Xcode, -DCMAKE_BUILD_TYPE=<Type> is not needed, and the build type is configured on compilation.

   Therefore, this cmake build command is used:

   cmake --build . --config Release --target benchmarks
   

   Options for --config are Release and Debug.
   Options for --target are benchmarks and tests

   Note

   With the Option -D toggle options can be activated supported only by this CMakeLists.txt. Available options are:

   Option	Description
   SAVE_JITS_TO_FILE	Saves the jitted kernels into a file if activated.

   To activate an option, the following must be written:

   cmake .. -D <Option>=ON
   

   For Example:

   cmake .. -D SAVE_JITS_TO_FILE=ON
   

6. Now we can build the project. The most desired command might be

   Options for --target are benchmarks and tests

   Option	Description
   benchmarks	Build the benchmarks executable.
   tests	Builds the tests executable.

Running the Benchmarks & Tests

The executables have been build in to the ../machine-learning-compilers/build directory with their corresponding name. E.g. If tests is built than the executable name is tests, for benchmarks the executable name would be benchmarks, etc.

All the executables can be found in ../machine-learning-compilers/build. The available executables are benchmarks and tests.

Note

They are available when build with their respective --target

E.g. the benchmarks executable can be run with the following command:

./benchmarks

The most desired command for the benchmarks might be:

./benchmarks --benchmark_counters_tabular=true --benchmark_repetitions=10 --benchmark_report_aggregates_only=true