hipfile 0.2.0.dev1

Python bindings for hipFile — direct-to-GPU file IO via AMD Infinity Storage (AIS)

hipFile Python Bindings

Python bindings for the hipFile C library.

[!CAUTION] These bindings in particular are experimental and the API will change.

Building & Installing

1. Setup a Python virtual environment.

   $ python3 -m venv .venv
   

2. Activate the Python virtual environment.

   $ source .venv/bin/activate
   

3. Build or install the C hipFile library. See INSTALL.md.
4. Build the hipFile package.
   and shared library using its configured search paths, including the source ../include location for headers, a CMake build-directory hint for the shared library, and the default ROCm install location. The search path can be overridden by setting HIPFILE_INCLUDE_DIR, HIPFILE_LIBRARY, or HIP_INCLUDE_DIR via the build arg -Ccmake.define.<KEY>=<VALUE>.

   (.venv) $ cd projects/hipfile/python
   (.venv) $ python -m build --wheel
   

   Or to install an editable version of the Python package:

   (.venv) $ cd projects/hipfile/python
   (.venv) $ pip install -e .
   

An editable Python package is a development package that allows for live changes to the Python source code to be applied immediately without needing to rebuild or reinstall the Python package under development. Any changes to the Cython source code will require a rebuild step.