oneDNN/doc/build/build.md

Build from Source

Download the Source Code

Download oneDNN source code or clone the repository.

git clone https://github.com/uxlfoundation/oneDNN.git

Build the Library

Set up the environment for the compiler

Ensure that all software dependencies are in place and have at least the minimal supported version.

Generate the build system

The oneDNN build system is based on CMake. Use the following command to generate a build system:

cmake -B <path-to-build> [-S <path-to-source>] [<options>]

In most cases, both -B and -S options are skipped with the assumption that <path-to-build> is the current folder and <path-to-source> is the one higher in the tree:

cd <path-to-onednn-source>
mkdir -p build ; cd build
cmake .. [<options>]

The following are a few useful options defined by CMake:

• G to specify build system generator (e.g. "Visual Studio 17 2022", Ninja, "Unix Makefiles").

• CMAKE_INSTALL_PREFIX to control the library installation location.

• CMAKE_BUILD_TYPE to select between build type (Release, Debug, RelWithDebInfo).

• CMAKE_PREFIX_PATH to specify directories to be searched for the dependencies located at non-standard locations.

See @ref dev_guide_build_options for detailed description of build-time configuration options defined by oneDNN.

Build the library

CMake provides a unified method for building a project, independent of the generator or operating system used:

cmake --build <path-to-build> [<options>]

Full list of options can be found here.

Linux/macOS

GCC, Clang, or Intel oneAPI DPC++/C++ Compiler

• Set up the environment for the compiler

# Uncomment the following lines to build with GCC
# export CC=gcc
# export CXX=g++

# Uncomment the following lines to build with Clang
# export CC=clang
# export CXX=clang++

# Uncomment the following lines to build with Intel oneAPI DPC++/C++ Compiler (x64 only)
# export CC=icx
# export CXX=icpx

• Generate the build system

mkdir -p build ; cd build
cmake ..

• Build the library

# Some generators, like Unix Makefiles, might default to single-threaded
# compilation. Parallelization can be controlled with:
cmake --build . --parallel $(nproc)

Intel oneAPI DPC++/C++ Compiler with SYCL runtime

• Set up the environment for the compiler

Intel oneAPI DPC++/C++ Compiler uses the setvars.sh script to set all required variables. The command below assumes you installed to the default folder. If you customized the installation folder, setvars.sh (Linux/macOS) is in your custom folder.

source /opt/intel/oneapi/setvars.sh

# Set Intel oneAPI DPC++/C++ Compiler as default C and C++ compilers
export CC=icx
export CXX=icpx

• Generate the build system

mkdir -p build ; cd build
cmake .. -DDNNL_CPU_RUNTIME=SYCL \
         -DDNNL_GPU_RUNTIME=SYCL

@note Open-source version of oneAPI DPC++ Compiler does not have the icx driver, use clang/clang++ instead. Open-source version of oneAPI DPC++ Compiler may not contain OpenCL runtime. In this case, you can use OPENCLROOT CMake option or environment variable of the same name to specify path to the OpenCL runtime if it is installed in a custom location.

• Build the library

cmake --build . --parallel $(nproc)

GCC targeting AArch64 on x64 host

• Set up the environment for the compiler

export CC=aarch64-linux-gnu-gcc
export CXX=aarch64-linux-gnu-g++

• Generate the build system

mkdir -p build ; cd build
cmake .. -DCMAKE_SYSTEM_NAME=Linux \
         -DCMAKE_SYSTEM_PROCESSOR=AARCH64 \
         -DCMAKE_LIBRARY_PATH=/usr/aarch64-linux-gnu/lib

• Build the library

cmake --build . --parallel $(nproc)

GCC with Arm Compute Library (ACL) on AArch64 host

• Set up the environment for the compiler

Download Arm Compute Library or build it from source and set ACL_ROOT_DIR to directory where it is installed.

export ACL_ROOT_DIR=<path/to/ComputeLibrary>
export CC=gcc
export CXX=g++

• Generate the build system

mkdir -p build ; cd build
cmake .. -DDNNL_AARCH64_USE_ACL=ON

• Build the library

cmake --build . --parallel $(nproc)

Windows

Microsoft Visual C++ Compiler

• Set up the environment for the compiler

Microsoft Visual Studio uses the VsDevCmd.bat script to set all required variables. The command below assumes you installed to the default folder. If you customized the installation folder, VsDevCmd.bat is in your custom folder.

"C:\Program Files\Microsoft Visual Studio\2022\Professional\Common7\Tools\VsDevCmd.bat" -startdir=none -arch=x64 -host_arch=x64

or open x64 Native Tools Command Prompt from start menu instead.

• Generate the build system

mkdir build
cd build
cmake .. -G "Visual Studio 17 2022"

• Build the library

cmake --build . --config=Release

@note Currently, the oneDNN build system has limited support for multi-config generators. Build configuration is based on the CMAKE_BUILD_TYPE option (Release by default), and CMake must be rerun from scratch every time the build type changes to apply the new build configuration. You can choose a specific build type with the --config option (the solution file supports both Debug and Release builds), but it must refer to the same build type (Release, Debug, etc.) as selected with the CMAKE_BUILD_TYPE option.

@note You can also open oneDNN.sln to build the project from the Microsoft Visual Studio IDE.

Intel oneAPI DPC++/C++ Compiler with SYCL Runtime

• Set up the environment for the compiler

Intel oneAPI DPC++/C++ Compiler uses the setvars.bat script to set all required variables. The command below assumes you installed to the default folder. If you customized the installation folder, setvars.bat is in your custom folder.

"C:\Program Files (x86)\Intel\oneAPI\setvars.bat"

:: Set Intel oneAPI DPC++/C++ Compiler as default C and C++ compilers
set CC=icx
set CXX=icx

or open Intel oneAPI Commmand Prompt from start menu instead.

• Generate the build system

mkdir build
cd build

cmake .. -G Ninja ^
         -DDNNL_CPU_RUNTIME=SYCL ^
         -DDNNL_GPU_RUNTIME=SYCL

@warning Intel oneAPI DPC++/C++ Compiler on Windows requires CMake v3.23 or later.

@warning Intel oneAPI DPC++/C++ Compiler does not support CMake's Microsoft Visual Studio generator.

@note Open-source version of oneAPI DPC++ Compiler does not have the icx driver, use clang/clang++ instead. Open-source version of oneAPI DPC++ Compiler may not contain OpenCL runtime. In this case, you can use OPENCLROOT CMake option or environment variable of the same name to specify path to the OpenCL runtime if it is installed in a custom location.

• Build the library

cmake --build .

Validate the Build

After building the library, you can run a predefined test set using:

ctest

The ONEDNN_TEST_SET build option set during the build configuration determines determines the scope and depth of the test set. Useful values are SMOKE (smallest set), CI (default), and NIGHTLY (most comprehensive). The test set can be reconfigured after the entire project has been built, and only the missing tests will be compiled.

cmake .. -DONEDNN_TEST_SET=NIGHTLY
cmake --build .
ctest

ctest supports filtering the test set by using the -R option. For example, to run only the GPU tests, use:

ctest -R gpu

Another useful option is --output-on-failure, which will print verbose output in case a test fails. Full set of options can be found here.

Build documentation

• Install the requirements

conda env create -f ../doc/environment.yml
conda activate onednn-doc

• Build the documentation

cmake --build . --target doc

Install library

Install the library, headers, and documentation

cmake --build . --target install

The install directory is specified by the CMAKE_INSTALL_PREFIX cmake variable. When installing in the default directory, the above command needs to be run with administrative privileges using sudo on Linux/Mac or a command prompt run as administrator on Windows.