Installation

The framework does not need to be explicitly installed in the python path. However, we strongly recommend using a python virtual environment to install all the dependencies.

You can use the pyproject.toml and pip to install the package:

# Install dependencies with its pyproject.toml configuration
pip install --deps-only .

# Install in development mode
pip install -e .

Other packages

For testing, there is a test_basic_flow.py in the test directory that makes use of a profile_util package. This is a MPI+OpenMP+CUDA/HIP C++ code that requires

• CMake (>3.20)

• C++ (c++17)

• OpenMP (optional)

• CUDA or HIP (optional)

• MPI (optional)

The package can be built running the build_cuda.sh and build_hip.sh scripts provided in the profile_util package. These use cmake to build a GPU-enabled, OpenMP-enabled code that can be used to test GPU and CPU flows.