NVIDIA/cutile-python

cuTile Python

cuTile Python is a programming language for NVIDIA GPUs. The official documentation can be found on docs.nvidia.com, or built from source located in the docs folder.

Installing from PyPI

cuTile Python is published on PyPI under the cuda-tile package name and can be installed with pip:

1

pip install cuda-tile

Currently, the CUDA Toolkit 13.1+ is required and needs to be installed separately.

Building from Source

cuTile is written mostly in Python, but includes a C++ extension which needs to be built. You will need:

A C++17-capable compiler, such as GNU C++ or MSVC;
CMake 3.18+;
GNU Make on Linux or msbuild on Windows;
Python 3.10+ with development headers (venv module is recommended but optional);
CUDA Toolkit 13.1+

On an Ubuntu system, the first four dependencies can be installed with APT:

1

sudo apt-get update && sudo apt-get install build-essential cmake python3-dev python3-venv

The CMakeLists.txt script will also automatically download the DLPack dependency from GitHub. If you wish to disable this behavior and provide your own copy of DLPack, set the CUDA_TILE_CMAKE_DLPACK_PATH environment variable to a local path to the DLPack source tree.

Unless you are already using a Python virtual environment, it is recommended to create one in order to avoid installing cuTile globally:

1
2


python3 -m venv env
source env/bin/activate

Once the build dependencies are in place, the simplest way to build cuTile is to install it in editable mode by running the following command in the source root directory:

1

pip install -e .

This will create the build directory and invoke the CMake-based build process. In editable mode, the compiled extension module will be placed in the build directory, and then a symbolic link to it will be created in the source directory. This makes sure that the pip install -e . command above is needed only once, and recompiling the extension after making changes to the C++ code can be done with make -C build which is much faster. This logic is defined in setup.py.

Running Tests

cuTile uses the pytest framework for testing. Tests have extra dependencies, such as PyTorch, which can be installed with

1

pip install -r test/requirements.txt

The tests are located in the test/ directory. To run a specific test file, for example test_copy.py, use the following command:

1

pytest test/test_copy.py

Copyright and License Information

cuTile-Python is under Apache 2.0 license. See the LICENSES folder for the full license text.