About

Simple real-time fluid simulations based on Jos Stam paper "Stable fluids". Implementation contains CPU version as well as GPU accelerated one.

Building

This application has following dependencies:

• CUDA
• Raylib

Note on Raylib - this library was used simply for the fact how easy it is to setup simple window and access underlying texture of the window (both for CPU and GPU). Since we want to avoid unnecesery CPU and GPU transactions over PCIe fabric during simulation on GPU, we want to access underlying window texture that is stored directly on the GPU. Luckily since Raylib is a simple wrapper around OpenGL this can be done as CUDA already is shipped with OpenGL interop functionalities. Now GPU solver, will directlly render into OpenGL texture created by Raylib context, in GPU code only OpenGL API is used. This is why need to link against OpenGL while compiling GPU code into a static library.

Note on building static library from CUDA - NVCC compiler will compile all functions marked with __host__ attribute using host native compiler (MSVC compiler on Windows), and use CUDA compiler for functions marked with __device__ attribute. Functions marked __host__ attribute will land in normal code sections, meanwhile __device__ code after being compiled to PTX format will be placed in .nvfatbin section (during runtime data in this section will be compiled by runtime/driver into compatible device code). Host code requires some OpenGL definitions but those are shipped with MSVC compiler already and include path and path to libraries should be resolved automatically by NVCC compiler during installation (NVCC will use MSVC resources).

[https://hpcgpu.mini.pw.edu.pl/cuda-compilation-toolchain/]

Instaling dependencies

1. MSVC is automatically installed with Visual Studio. This instalation should already contain OpenGL header files and static libraries to link against.
2. Using CUDA toolkit instalator for Windows will already install appropriate version for MSVC compiler since host side code for CUDA projects is already compiled by host native compiler (which is MSVC in this case)
3. Raylib 5.5 can be installed from official Github realese: [https://github.com/raysan5/raylib/releases/tag/5.5] Select version appropriate for the compiler, for Windows with MSVC compiler this would be:

Building scripts

Commands needed to compile the project are present in build_cpu.bat and build_gpu.bat. There is also one joined script that will call both of those. To use this script fill out path to unzipped Raylib directory in build.bat file. Launch x64 Native Tools CMD (it is a CMD shipped with Visual Studio that sets some enviroment variables for building - this step is important and if skipped build script will not work). Change working directory to the project directory and launch build.bat script.

Minimal step to build project

Those steps assume Visual Studio and CUDA already installed.

1. Download and unzip Raylib realese for MSVC compiler (x64 version)
2. Modify path to Raylib resources in the build.bat script
3. Launch x64 native tools console (shipped with MSVC compiler)
4. Change working directory to the project directory
5. Run build script