Skip to content
shaia edited this page Oct 5, 2026 · 4 revisions

CFD Library

Welcome to the CFD Library wiki! This is a 2D/3D Computational Fluid Dynamics framework implemented in C. These pages describe version 0.4.0; see the CHANGELOG for what changed in each release.

Quick Links

Features

  • 2D/3D structured grids - Uniform and stretched grid generation
  • Multiple solvers - Explicit Euler, Projection Method, RK2 (Heun's method), RK4 (classical), with scalar, SIMD (AVX2), OpenMP, and GPU variants
  • Linear solvers - Jacobi, SOR, Red-Black SOR, CG/PCG, BiCGSTAB, GMRES(m), and geometric multigrid (also as a CG preconditioner and as the projection pressure solver)
  • Discretization options - Central or first-order upwind convection; explicit or implicit (backward Euler, Crank-Nicolson) viscous time stepping
  • Heat transfer - Energy equation with Boussinesq buoyancy and thermal boundary conditions
  • RANS turbulence - k-epsilon and Spalart-Allmaras with wall functions (2D uniform grids; scalar, OpenMP, and AVX2 backends)
  • Checkpoint/restart - Versioned binary .cfdchk files that save and restore the full simulation state
  • Neural-network inference - Pure-C evaluator for small .cfdnn models, used for learned turbulence closures
  • Flexible output - VTK for visualization, CSV for data analysis
  • Pluggable architecture - Easy to add new solvers
  • Cross-platform - Windows, Linux, macOS support

Minimal Example

#include "cfd/api/simulation_api.h"

int main() {
    // Create 100x100 2D simulation on unit square (nz=1 for 2D)
    simulation_data* sim = init_simulation(100, 100, 1, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0);

    // Register VTK output every 10 steps
    simulation_register_output(sim, OUTPUT_VELOCITY_MAGNITUDE, 10, "velocity");

    // Run 100 time steps
    for (int step = 0; step < 100; step++) {
        run_simulation_step(sim);
        simulation_write_outputs(sim, step);
    }

    free_simulation(sim);
    return 0;
}

Available Solvers

Solver Type Best For
explicit_euler CPU Learning, debugging
explicit_euler_optimized CPU+SIMD (AVX2) Large grids on modern CPUs
explicit_euler_omp CPU+OpenMP Multi-core CPU acceleration
explicit_euler_gpu GPU Very large grids with NVIDIA GPU
projection CPU Accurate pressure-velocity coupling
projection_optimized CPU+SIMD (AVX2) Production simulations
projection_omp CPU+OpenMP Multi-core production simulations
projection_gpu GPU High-performance GPU simulations
rk2 CPU Second-order temporal accuracy
rk2_optimized CPU+SIMD (AVX2) High-accuracy large grids
rk2_omp CPU+OpenMP High-accuracy multi-core
rk2_gpu GPU High-accuracy on NVIDIA GPU
rk4 CPU Fourth-order temporal accuracy
rk4_optimized CPU+SIMD (AVX2) Fourth-order accuracy, large grids
rk4_omp CPU+OpenMP Fourth-order accuracy, multi-core
rk4_gpu GPU Fourth-order accuracy on NVIDIA GPU

The *_optimized solvers are AVX2 only (there are no NEON Navier-Stokes kernels). They need a build with -DCFD_ENABLE_AVX2=ON (off by default) and a CPU with AVX2 and FMA; otherwise their init returns CFD_ERROR_UNSUPPORTED. The *_gpu solvers are registered only in a -DCFD_ENABLE_CUDA=ON build, and the *_omp solvers only when OpenMP is found.

Documentation

Getting Help

Clone this wiki locally