For my Computer Animation final project I implemented Position Based Fluids.

The main challenge of this assignment is optimizing the neighbor lookups. I used the paper they linked in Position Based Fluids to implement my acceleration structure, which is simply a uniform grid. Its an old Nvidia white paper on CUDA Particles.

I only finished a multithreaded CPU implementation before the class deadline, but just a day after I finished my GPU implementation using compute shaders to support massively more particles. There’s still room for optimization in it since I was crude with my synchronization to make sure it was not the reason for any of my issues during development.

Here is the video discussing the implementation, at this point I only had the CPU implementation working:

Libraries used:
Vulkan - Graphics API
VulkanMemoryAllocator - Easy buffer/image allocation
VkBootstrap - Reduce verbosity of startup
fastgltf - Model loading
glm - Math library
SDL3 - Windowing
Dear ImGui - Immediate mode GUI