Fluid Simulator (Rewrite WIP)

A Position Based Fluids/Particles simulator focused on CPU/CUDA parity, high-performance CUDA kernels, and a simple visualization path.

Goals

• CPU and CUDA parity for the PBF/PBD solver.
• High-performance CUDA implementation for large particle counts.
• Simple visualization (headless outputs or on-the-fly viewer).

Status

• Rewrite in progress; minimal build/run docs are below.
• See CODEX.md for rewrite guidance and parity expectations.

Docker Build

Image (CUDA base; CPU backend still works):

1. docker build -t fluidsim .
2. docker run --rm fluidsim ./build/fluidsim --backend=cpu

CUDA run (requires NVIDIA Container Toolkit + compatible driver): docker run --rm --gpus all fluidsim ./build/fluidsim --backend=cuda

Example with scene + output volume: docker run --rm --gpus all -v "$PWD/output_docker_cuda:/opt/fluidsim/output_cuda" fluidsim ./build/fluidsim --backend=cuda --scene scene/fluid_million.json --steps-per-sec 120 --fps 30 --duration 4 --enable-scorr --enable-xsph --plane-restitution 0.05 --plane-friction 0.1 --output-dir output_cuda --debug-print

Legacy

• The previous implementation (CS 284A Spring 2020 final project) moved to the legacy branch, see also: legacy README.