Fluids

A high-performance fluid simulation project for the Multicore 2023/24 course.

1. Project Overview

This project simulates incompressible fluid dynamics using Eulerian grids and a numerical solver. The implementation supports three execution modes: Serial (CPU), OpenMP (Multithreaded CPU), and CUDA (GPU-accelerated).

Graphics

The visualization is powered by OpenGL, using custom vertex and fragment shaders to render the simulation efficiently. The rendering pipeline includes:

• GLFW for window and input management.
• GLEW to handle OpenGL extensions.
• Dear ImGui for real-time parameter tuning.

💡 [Insert shader code snippet or diagram here]

Physics

The simulation is based on Jos Stam’s Stable Fluids method, ensuring numerical stability in real-time applications. The main steps include:

• Advection: Moves the fluid according to its velocity field, solved via a semi-Lagrangian method.
• Diffusion: Spreads velocity values using implicit diffusion, solved with a linear system.
• Projection: Ensures the velocity field is divergence-free using a pressure correction step.

📜 Reference: Real-Time Fluid Dynamics for Games (Jos Stam, 2003)

Implementation

This project is implemented in three different ways:

• Serial (CPU-only): Uses a naive Gauss-Seidel solver.
• OpenMP (Multicore CPU): Switches to a Jacobi solver for parallelization.
• CUDA (GPU-accelerated): Optimized with memory management to reduce RAM-VRAM transfers.

💡 [Insert flowchart or pseudocode for implementation comparison]

---

2. Running the Simulator

Dependencies

Ensure you have the following installed:

• GLFW
• OpenGL
• GLEW
• OpenMP (for CPU parallelization)
• CUDA Toolkit (for GPU acceleration)

Building the Project

mkdir build && cd build
cmake ..
make -j$(nproc)

Running the Simulator

./fluids_sim

Settings and Visualization Modes

The simulation offers different visualization options:

• Diffusion Mode: Shows how particles spread over time.
• Velocity Field: Visualizes the motion of the fluid.
• Vorticity Map: Highlights areas of rotational motion.

💡 [Leave space for screenshots of each mode]

Keybindings

• R - Reset simulation
• P - Pause/Resume simulation
• +/- - Increase/Decrease timestep

💡 [Provide a complete list of controls]

---

3. Performance & Scaling

This section analyzes performance across different implementations, focusing on:

• Speedup from OpenMP and CUDA over Serial
• VRAM vs RAM bandwidth optimizations
• Computational cost of Jacobi vs Gauss-Seidel solvers

📊 [Leave space for benchmark graphs comparing Serial, OpenMP, and CUDA]

---

🚀 Stay tuned for updates! Feel free to contribute or report issues.