XPG

XPG is a library for writing native cross-platform real-time graphics applications and tools with minimal dependencies. It is based on Vulkan for performance and portability. It provides helpers to simplify application development but exposes Vulkan concepts, functions, and types directly to give full control to applications and allow access all functionality and extensions that Vulkan offers.

XPG is written in a very C-like like C++, it makes lightweight use of templates and destructors for a few simple data structures, and it uses operator overloading for bounds checks and vector math.

Features

• Helpers for device initialization, (optionally) window management, and most of the Vulkan API.
• ImGui integration based on the GLFW and Vulkan backends.
• Shader compilation and reflection utilities based on Slang or glslang and SPIR-V tools.
• First-class Python bindings for graphics, imgui and shader compilation and reflection for Python 3.8+.

Project structure

The project currently contains 3 main components:

• XPG: C++ library. Source is in src/include/ and src/lib/. Example applications are in apps/ and shaders in apps/shaders/.
• PyXPG: Python bindings for XPG using nanobind. Source of the bindings is in src/python/ and examples are in python/.
• Ambra: Ambra is a pure-Python 3D viewer based on PyXPG. Source is in ambra/ and examples in ambra/examples/. Ambra is meant as a standalone tool and is one of the main motivations driving the design of both XPG and PyXPG.

Dependencies

All dependencies are included as git submodules for versioning and for self-contained builds and source distributions:

• Vulkan SDK components (tracking the latest official release).
• GLFW for cross-platform window creation and input collection.
• ImGui and ImPlot for UI and dear_bindings for imgui Python binding generation.
• glslang and slang for shader compilation and reflection.
• GLM for vector and matrix math.
• Nanobind for writing Python bindings.

On Linux, GLFW has a few build dependencies for X11 and Wayland support. The easiest way to install these is from your distribution package manager. See the Build section below for more details.

Build

Building the XPG library, Python bindings, and sample applications requires a C++20 compiler and cmake. The recommended way to build XPG applications is to build XPG as a static library, this will produce portable executables with no runtime dependencies other than the system standard libraries.

XPG can be built for any platform supported by the Vulkan SDK and GLFW. It's currently tested on x64 Windows, x64 and aarch64 Linux and aarch64 MacOS.

Building XPG does not require the Vulkan SDK to be installed. But, if available, XPG can enable and configure the validation layer at runtime.

CMake options

The CMake build can be configured with the following XPG-specific options:

• XPG_BUILD_APPS: If set to ON build example XPG applications in ./app. Default: ON.
• XPG_BUILD_PYTHON: If set to ON build the pyxpg Python module. Requires Python header files to be installed and discoverable by find_package. Default: ON.
• XPG_BUILD_SLANG: If set to ON build slang and the pyxpg.slang Python module. Default: ON.
• XPG_BUILD_GLSLANG_AND_SPIRV: If set to ON build glslang and spirv tools (experimental). Deafult: OFF.
• XPG_MSVC_ANALYZE: If set to ON and compiling with MSVC, build with the /analyze flag set for additional checks (significantly increases compilation time). Default: OFF.
• XPG_MOLTENVK_PATH: If specified, link statically with the MoltenVK library pointed to this path. Otherwise Vulkan is loaded dynamically at runtime by volk.

Most dependencies also use CMake and their variables can also be specified at configuration time. By default, XPG configures some of the dependencies to disable optional features and configure them for static linking.

Windows

A C++ compiler and cmake can be installed with the C/C++ development package from the Visual Studio installer. XPG compilation on Windows is tested with both MSVC and Clang. For building PyXPG Python must be installed system-wide along with development header files.

Linux

XPG is tested with both Clang and GCC on x64 and aarch64 Ubuntu 24.04. Other distributions are expected to work by installing the respective packages. XPG builds GLFW with support for X11, Wayland, or both, depending on which packages are found at CMake configuration time.

The following packages are required for building XPG with X11 support:

sudo apt install libx11-dev libxrandr-dev libxinerama-dev libxcursor-dev libxi-dev

The following packages are required for building XPG with Wayland support:

sudo apt install libwayland-dev libxkbcommon-dev wayland-protocols

The following packages are required for building PyXPG:

sudo apt install libpython3-dev

PyXPG Python wheels for distribution are built on manylinux_2_28 and musllinux_1_2. See pyproject.toml for the list of build dependencies on those platforms.

MacOS

On MacOS XPG is tested using Clang on aarch64 MacOS 14. Intel MacOS should also work but is not regularly tested.

MoltenVK is required for Vulkan support. By default, MoltenVK is expected to be installed by the Vulkan SDK and dynamically loaded at runtime by volk. Alternatively, MoltenVK can be linked statically to simplify distribution of binaries by specifying the XPG_MOLTENVK_PATH CMake option.

Example build commands

Full build

cmake -B build/
cmake --build build/

Minimal build (no apps, no Python bindings, and no slang)

cmake -B build/ -DXPG_BUILD_APPS=OFF -DXPG_BUILD_PYTHON=OFF -DXPG_BUILD_SLANG=OFF
cmake --build build/

Editable Python build

For development of PyXPG it can be convenient to setup an editable build that will check for changes and recompile the bindings whenever the pyxpg module is imported.

This can be achieved with the following command:

pip install scikit-build-core
pip install -C cmake.build-type=Debug -C editable.rebuild=true  --no-build-isolation -ve .

See the scikit-build-core documentation for more details on build configuration.

Ambra

Ambra is a pure-Python library based on PyXPG for creating 3D visualization and GUI tools. Ambra aims to be a tool that can be quickly installed in most environments and can bring any data from any source to the screen in as few lines of code as possible.

The main motivation for writing Ambra in Python is to minimize the effort and time required for setup and customizations. Compared to other 3D tools, Ambra is designed to allow full customization and configuration for the application needs. You can start using Ambra with some of the default primitives, UIs and controls, but as your application grows, you are encouraged to customize any big or small part of the library. For example, you can add your own primitives, shaders, render passes, data streaming systems, etc..

Even though Python isn't the fastest language, Ambra tries to provide fast primitives that limit the overhead of the interpreter. This includes supporting modern GPU features like asynchronous streaming, GPU driven rendering, and bindless resources.

Ambra has minimal dependencies and should be easy to integrate into any Python 3.8+ environment alongside other packages.

Setup

Ambra is deep in development and Python wheels are not yet available. Currently ambra does not depend explicitly on pyxpg in its pyproject.toml to avoid downloading wheels from Pypi when installed.

The recommended way to install Ambra is to clone the repository and install from source both pyxpg and ambra to ensure the latest version of both is in use.

Example setup:

git clone --recursive https://github.com/ramenguy99/xpg.git
cd xpg
pip install scikit-build-core
pip install -C cmake.build-type=Debug -C editable.rebuild=true  --no-build-isolation -ve .
cd ambra
pip install -e .

This will install pyxpg in debug and editable mode and ambra in editable mode. After the setup you can run the example scripts in ambra/examples/ to test the installation.