A minimalist wrapper around the Vulkan API to write easy, portable, and (somewhat) safe compute shaders in C!
I wanted a compute shader to parallelise one of my projects, it was meant to run on a super-computer, so I needed something portable, fast, and able to run without a window, so Vulkan was the best choice. Sadly I'm not very experienced with Vulkan, and I would've had a hard time transferring my whole algorithm to it. So I made this side project that automatically manages variables, memory, and the pipeline, so I could focus on the algorithm itself.
- Very easy to use, and doesn't require much code to get started. I would say it's about as hard as CUDA, so quite easy if you have a basic understanding of parallel programming.
- Fast & Portable, because so is Vulkan.
- Error handling and checking, if you want, there are automatic memory checks, that you can enable or disable, which prevents basic mistakes like writing to a variable that doesn't exist, or not finding the shader file, or Vulkan not being supported. All of that without needing to mess with Validation Layers.
- Can also be used with C++
- No need to create a window, or use a library like GLFW, it's meant to be used without a window, and it's very lightweight.
- Can work with any shader language you like, I chose GLSL in the examples simply because it's the closest to C in terms of syntax
- Not very flexible, as it's meant to be simple, it doesn't support many features of Vulkan, and it's not meant to be used for complex graphics, only for compute shaders.
If all you need is to send data to the GPU, make it do some calculations, and get the results back, then it's perfect. For example : doing operations on very large arrays, matrices and so on... However, if you need to do more complex operations, you should use Vulkan directly, or a wrapper with more features.
A basic program will look like this :
- Step 1: Init the Culkan instance with
culkanInit(), which takes a shader binding layout, and a shader file. - Step 2: Write to the variables with
culkanWriteBinding()orculkanWriteGPUVariable(). - Step 3: Finish the setup with
culkanSetup(). - Step 4: Run the shader with
culkanRun(). - Step 5: Read the results with
culkanReadBinding()orculkanReadGPUVariable(). - Step 6: Free the resources with
culkanDestroy().
So it's quite simple, and definitely less verbose than Vulkan.
- A C or C++ compiler
- Vulkan libraries
- GPU drivers that support Vulkan
You should be able to run on any architecture/OS/distro with these requirements, I tested it on Linux with an AMD GPU (Manjaro, AMD Lucienne).
I am a bit stupid, so if the code is a bit stupid, that's why. Clang tidy is mad at me for a few reasons, I'll have to fix that. FIXME The casing and naming style may change in the future, because I'm not sure on if I want to follow Vulkan's naming style or dissociate from it, but it shouldn't be too bad to change it in your code if you ever use this. I promise I will do better documentation and examples in the future, and clean up some stuff. But I don't want to be too off-schedule with my main project. I learned a bit about Vulkan, so that's cool. If this is useful to you, then I'm happy :D