hephaistos 0.9.0

GPU computing framework using the Vulkan API

Hephaistos

Hephaistos is a framework for creating GPGPU appllication running on top of the Vulkan API, allowing it to run on all major OS and on a wide range of hardware such as all modern AMD, Intel and NVIDIA GPUs.

The library aims to reduce the amount of boiler plate code, albeit in doing so it might hide optimization possibility Vulkan provides.

Dependency

To ease the building process, all external dependency are stored inside the external directory. Check there for more information.

Mac OS

Mac OS does not support Vulkan on its own, but requires a translation layer called MoltenVK and can be installed for example via Homebrew:

brew install molten-vk

Since future updates to MoltenVK may enable more features without rebuilding hephaistos, it is not included in the library itself.

Install

There exist pre-built python packages on PyPI and may be easily installed via pip:

pip install hephaistos

For some rare platforms you might need to build from sources.

There are also no pre-built libraries for C++, as the API depends on the STL and thus has no stable ABI (i.e. you would need to create a build for each compiler).

Building

C++

For building the C++ library you require:

• C++20 compatible compiler
• CMake at least version 3.18

The dependencies are located in the external directory. In there is a readme file for more information.

The debug build requires the validation layers, which are part of the Vulkan SDK.

Hephaistos expects the kernel/shader code to be already compiled into SPIR-V bytecode, so you must likely also need an compiler for that. The most popular ones are:

• glslangvalidator
• glslc

See the readme inside the examples folder for a hint on how to use them.

Python

There are also python bindings using nanobind and scikit-build.

To build the python package you can simply run from the project's root directory:

pip install .

Note that the binding relies on the C++ library and thus has the same dependencies. The dependencies on the Python side are handled by pip.

Learning

There are some examples provided in the examples directory for C++ and in notebooks for Python.

License

Hephaistos is licensed under the MIT license. See LICENSE. External libraries are distributed under their own license. The ones located in external are always acompanied by a copy of their license.