CFD Python

Python bindings for high-performance CFD simulation library using CPython C-API with stable ABI (abi3).

Features

• High Performance: Direct bindings to optimized C library with SIMD (AVX2/NEON) and OpenMP support
• Stable ABI: Compatible across Python 3.9+ versions
• Multiple Backends: Scalar, SIMD, OpenMP, and CUDA (GPU) backends with runtime detection
• Boundary Conditions: Full BC support (Neumann, Dirichlet, no-slip, inlet, outlet)
• Dynamic Solver Discovery: New solvers automatically available
• Multiple Output Formats: VTK and CSV export support
• Error Handling: Rich exception hierarchy with detailed error messages

Installation

Installing uv (optional but recommended)

uv is a fast Python package manager. Install it with:

# On macOS/Linux
curl -LsSf https://astral.sh/uv/install.sh | sh

# On Windows
powershell -ExecutionPolicy ByPass -c "irm https://astral.sh/uv/install.ps1 | iex"

# Or with pip
pip install uv

From PyPI

Using uv (recommended):

uv pip install cfd-python

Or with pip:

pip install cfd-python

From Source

The Python package requires the C CFD library to be built first. By default, it expects the library at ../cfd relative to the cfd-python directory. You can override this by setting the CFD_ROOT environment variable.

1. Build the C CFD library (static):

   cd ../cfd
   cmake -B build -DCMAKE_BUILD_TYPE=Release -DBUILD_SHARED_LIBS=OFF
   cmake --build build --config Release
   

2. Install the Python package:

   cd ../cfd-python
   pip install .
   

   Or with uv for faster installs:

   uv pip install .
   

   With a custom library location:

   CFD_ROOT=/path/to/cfd pip install .
   

For Development

We use uv for fast dependency management:

# Create virtual environment
uv venv

# Activate it
source .venv/bin/activate      # Linux/macOS
source .venv/Scripts/activate  # Windows (Git Bash)
.venv\Scripts\activate         # Windows (cmd)

# Install with dev dependencies
uv pip install -e ".[test,dev]"

Alternatively, using pip:

pip install -e ".[test,dev]"

Quick Start

import cfd_python

# List available solvers
print(cfd_python.list_solvers())
# ['explicit_euler', 'explicit_euler_optimized', 'projection', ...]

# Run a simple simulation
velocity_magnitude = cfd_python.run_simulation(50, 50, steps=100)
print(f"Computed {len(velocity_magnitude)} velocity values")

# Create a grid
grid = cfd_python.create_grid(100, 100, 0.0, 1.0, 0.0, 1.0)
print(f"Grid: {grid['nx']}x{grid['ny']}")

# Get default solver parameters
params = cfd_python.get_default_solver_params()
print(f"Default dt: {params['dt']}")

API Reference

Simulation Functions

run_simulation(nx, ny, steps=100, xmin=0.0, xmax=1.0, ymin=0.0, ymax=1.0, solver_type=None, output_file=None)

Run a complete simulation with default parameters.

Parameters:

• nx, ny: Grid dimensions
• steps: Number of time steps (default: 100)
• xmin, xmax, ymin, ymax: Domain bounds (optional)
• solver_type: Solver name string (optional, uses library default)
• output_file: VTK output file path (optional)

Returns: List of velocity magnitude values

run_simulation_with_params(nx, ny, xmin, xmax, ymin, ymax, steps=1, dt=0.001, cfl=0.2, solver_type=None, output_file=None)

Run simulation with custom parameters and solver selection.

Parameters:

• nx, ny: Grid dimensions
• xmin, xmax, ymin, ymax: Domain bounds
• steps: Number of time steps (default: 1)
• dt: Time step size (default: 0.001)
• cfl: CFL number (default: 0.2)
• solver_type: Solver name string (optional, uses library default)
• output_file: VTK output file path (optional)

Returns: Dictionary with velocity_magnitude, nx, ny, steps, solver_name, solver_description, and stats

create_grid(nx, ny, xmin, xmax, ymin, ymax)

Create a computational grid.

Parameters:

• nx, ny: Grid dimensions
• xmin, xmax, ymin, ymax: Domain bounds

Returns: Dictionary with nx, ny, xmin, xmax, ymin, ymax, x_coords, y_coords

create_grid_stretched(nx, ny, xmin, xmax, ymin, ymax, beta)

Create a grid with non-uniform (stretched) spacing.

Parameters:

• nx, ny: Grid dimensions
• xmin, xmax, ymin, ymax: Domain bounds
• beta: Stretching parameter (higher = more clustering)

Returns: Dictionary with grid info including x_coords, y_coords

Note: The stretched grid implementation has a known bug - see ROADMAP.md for details.

get_default_solver_params()

Get default solver parameters.

Returns: Dictionary with keys: dt, cfl, gamma, mu, k, max_iter, tolerance

Solver Discovery

list_solvers()

Get list of all available solver names.

Returns: List of solver name strings

has_solver(name)

Check if a solver is available.

Returns: Boolean

get_solver_info(name)

Get information about a solver.

Returns: Dictionary with name, description, version, capabilities

Dynamic Solver Constants

Solver constants are automatically generated from the registry:

cfd_python.SOLVER_EXPLICIT_EULER           # 'explicit_euler'
cfd_python.SOLVER_EXPLICIT_EULER_OPTIMIZED # 'explicit_euler_optimized'
cfd_python.SOLVER_PROJECTION               # 'projection'
# ... more solvers as registered in C library

Backend Availability (v0.1.6+)

Query and select compute backends at runtime:

import cfd_python

# Check available backends
print(cfd_python.get_available_backends())
# ['Scalar', 'SIMD', 'OpenMP']  # CUDA if GPU available

# Check specific backend
if cfd_python.backend_is_available(cfd_python.BACKEND_SIMD):
    print("SIMD backend available!")

# Get backend name
print(cfd_python.backend_get_name(cfd_python.BACKEND_OMP))  # 'OpenMP'

# List solvers for a backend
omp_solvers = cfd_python.list_solvers_by_backend(cfd_python.BACKEND_OMP)
print(f"OpenMP solvers: {omp_solvers}")

Backend Constants:

• BACKEND_SCALAR: Basic scalar CPU implementation
• BACKEND_SIMD: SIMD-optimized (AVX2/NEON)
• BACKEND_OMP: OpenMP parallelized
• BACKEND_CUDA: CUDA GPU acceleration

Boundary Conditions

Apply various boundary conditions to flow fields:

import cfd_python

# Create velocity field (as flat lists)
nx, ny = 50, 50
u = [0.0] * (nx * ny)
v = [0.0] * (nx * ny)

# Apply uniform inlet on left edge
cfd_python.bc_apply_inlet_uniform(
    u, v, nx, ny,
    u_inlet=1.0, v_inlet=0.0,
    edge=cfd_python.BC_EDGE_LEFT
)

# Apply zero-gradient outlet on right edge
cfd_python.bc_apply_outlet_velocity(u, v, nx, ny, cfd_python.BC_EDGE_RIGHT)

# Apply no-slip walls on top and bottom
cfd_python.bc_apply_noslip(u, v, nx, ny)

# Check/set BC backend
print(f"BC Backend: {cfd_python.bc_get_backend_name()}")
if cfd_python.bc_backend_available(cfd_python.BC_BACKEND_OMP):
    cfd_python.bc_set_backend(cfd_python.BC_BACKEND_OMP)

BC Type Constants:

• BC_TYPE_PERIODIC: Periodic boundaries
• BC_TYPE_NEUMANN: Zero-gradient boundaries
• BC_TYPE_DIRICHLET: Fixed value boundaries
• BC_TYPE_NOSLIP: No-slip wall (zero velocity)
• BC_TYPE_INLET: Inlet velocity specification
• BC_TYPE_OUTLET: Outlet conditions

BC Edge Constants:

• BC_EDGE_LEFT, BC_EDGE_RIGHT, BC_EDGE_BOTTOM, BC_EDGE_TOP

BC Backend Constants:

• BC_BACKEND_AUTO: Auto-select best available
• BC_BACKEND_SCALAR: Single-threaded scalar
• BC_BACKEND_OMP: OpenMP parallel
• BC_BACKEND_SIMD: SIMD + OpenMP (AVX2/NEON)
• BC_BACKEND_CUDA: GPU acceleration

BC Functions:

• bc_apply_scalar(field, nx, ny, bc_type): Apply BC to scalar field
• bc_apply_velocity(u, v, nx, ny, bc_type): Apply BC to velocity fields
• bc_apply_dirichlet(field, nx, ny, left, right, bottom, top): Fixed values
• bc_apply_noslip(u, v, nx, ny): Zero velocity at walls
• bc_apply_inlet_uniform(u, v, nx, ny, u_inlet, v_inlet, edge): Uniform inlet
• bc_apply_inlet_parabolic(u, v, nx, ny, max_velocity, edge): Parabolic inlet
• bc_apply_outlet_scalar(field, nx, ny, edge): Zero-gradient outlet
• bc_apply_outlet_velocity(u, v, nx, ny, edge): Zero-gradient outlet

Derived Fields & Statistics

Compute derived quantities from flow fields:

import cfd_python

# Compute velocity magnitude
u = [1.0] * 100
v = [0.5] * 100
vel_mag = cfd_python.compute_velocity_magnitude(u, v, 10, 10)

# Calculate field statistics
stats = cfd_python.calculate_field_stats(vel_mag)
print(f"Min: {stats['min']}, Max: {stats['max']}, Avg: {stats['avg']}")

# Comprehensive flow statistics
p = [0.0] * 100  # pressure field
flow_stats = cfd_python.compute_flow_statistics(u, v, p, 10, 10)
print(f"Max velocity: {flow_stats['velocity_magnitude']['max']}")

CPU Features Detection

Detect SIMD capabilities at runtime:

import cfd_python

# Check SIMD architecture
arch = cfd_python.get_simd_arch()
name = cfd_python.get_simd_name()  # 'avx2', 'neon', or 'none'

# Check specific capabilities
if cfd_python.has_avx2():
    print("AVX2 available!")
elif cfd_python.has_neon():
    print("ARM NEON available!")

# General SIMD check
if cfd_python.has_simd():
    print(f"SIMD enabled: {name}")

SIMD Constants:

• SIMD_NONE: No SIMD support
• SIMD_AVX2: x86-64 AVX2
• SIMD_NEON: ARM NEON

Error Handling

Handle errors with Python exceptions:

import cfd_python
from cfd_python import (
    CFDError,
    CFDMemoryError,
    CFDInvalidError,
    CFDDivergedError,
    raise_for_status,
)

# Check status codes
status = cfd_python.get_last_status()
if status != cfd_python.CFD_SUCCESS:
    error_msg = cfd_python.get_last_error()
    print(f"Error: {error_msg}")

# Use raise_for_status helper
try:
    raise_for_status(status, context="simulation step")
except CFDDivergedError as e:
    print(f"Solver diverged: {e}")
except CFDInvalidError as e:
    print(f"Invalid parameter: {e}")
except CFDError as e:
    print(f"CFD error: {e}")

# Clear error state
cfd_python.clear_error()

Error Constants:

• CFD_SUCCESS: Operation successful (0)
• CFD_ERROR: Generic error (-1)
• CFD_ERROR_NOMEM: Out of memory (-2)
• CFD_ERROR_INVALID: Invalid argument (-3)
• CFD_ERROR_IO: File I/O error (-4)
• CFD_ERROR_UNSUPPORTED: Operation not supported (-5)
• CFD_ERROR_DIVERGED: Solver diverged (-6)
• CFD_ERROR_MAX_ITER: Max iterations reached (-7)

Exception Classes:

• CFDError: Base exception class
• CFDMemoryError(CFDError, MemoryError): Memory allocation failed
• CFDInvalidError(CFDError, ValueError): Invalid argument
• CFDIOError(CFDError, IOError): File I/O error
• CFDUnsupportedError(CFDError, NotImplementedError): Unsupported operation
• CFDDivergedError(CFDError): Solver diverged
• CFDMaxIterError(CFDError): Max iterations reached

Output Functions

set_output_dir(directory)

Set the output directory for VTK/CSV files.

write_vtk_scalar(filename, field_name, data, nx, ny, xmin, xmax, ymin, ymax)

Write scalar field to VTK file.

write_vtk_vector(filename, field_name, u_data, v_data, nx, ny, xmin, xmax, ymin, ymax)

Write vector field to VTK file.

write_csv_timeseries(filename, step, time, u_data, v_data, p_data, nx, ny, dt, iterations, create_new=False)

Write simulation timeseries data to CSV file.

Parameters:

• filename: Output file path
• step: Time step number
• time: Simulation time
• u_data, v_data, p_data: Flow field data as lists (size nx*ny)
• nx, ny: Grid dimensions
• dt: Time step size
• iterations: Number of solver iterations
• create_new: If True, create new file; if False, append

Output Type Constants

cfd_python.OUTPUT_VELOCITY_MAGNITUDE  # Velocity magnitude scalar (VTK)
cfd_python.OUTPUT_VELOCITY            # Velocity vector field (VTK)
cfd_python.OUTPUT_FULL_FIELD          # Complete flow field (VTK)
cfd_python.OUTPUT_CSV_TIMESERIES      # Time series (CSV)
cfd_python.OUTPUT_CSV_CENTERLINE      # Centerline profile (CSV)
cfd_python.OUTPUT_CSV_STATISTICS      # Global statistics (CSV)

Migration from v0.1.0

If you're upgrading from an older version, note these changes:

API Changes

1. Type names changed (internal, transparent to Python users)
2. Error handling improved: Use get_last_error(), get_last_status(), clear_error()
3. New solver backends: SIMD, OpenMP, CUDA available via backend_is_available()

New Features in v0.1.6

• Boundary conditions: Full BC API with inlet/outlet support
• Backend selection: Query and select compute backends at runtime
• Derived fields: compute_velocity_magnitude(), compute_flow_statistics()
• Error handling: Python exception classes with raise_for_status()
• CPU detection: has_avx2(), has_neon(), has_simd()

See MIGRATION_PLAN.md for detailed migration information.

Requirements

• Python 3.9+
• uv (recommended) or pip

For building from source:

• C compiler (MSVC on Windows, GCC/Clang on Unix)
• CMake 3.15+

License

MIT License