Cython链接共享库报错？Cuba库Cuhre例程调用问题

Got it, let's work through getting your custom libcuhre.so hooked up with Cython—since you ran into issues linking the static Cuba library directly, this shared library approach should smooth things out.

First, let's confirm your shared library is properly compiled (this is the foundation). You need to build cuhre.c with position-independent code (required for shared libraries) and link against the Cuba library to resolve the underlying Cuhre calls. Use a command like this:

gcc -fPIC -shared -o libcuhre.so cuhre.c -I/path/to/cuba/include -L/path/to/cuba/lib -lcuba

• Replace /path/to/cuba/include and /path/to/cuba/lib with the actual paths to your Cuba library's headers and binaries.
• The -lcuba flag tells the linker to pull in the Cuba library (dynamic is safer here, but static works if you have the right flags).

Step 1: Write the Cython Wrapper #

Create a file named cuhre_wrapper.pyx to bridge your C shared library with Python. We'll declare the C types/functions from cuhre.h and create a wrapper that lets Python functions act as the integrand:

# Declare the C types and functions from your cuhre.h
cdef extern from "cuhre.h":
    # Match the integrand_t typedef from Cuba's headers
    ctypedef void (*integrand_t)(int ndim, const double xx[], int ncomp, double ff[], void *userdata)
    double tryCuhre(integrand_t t, void *ud)

# Intermediate Cython function to wrap Python integrands
cdef void py_integrand(int ndim, const double xx[], int ncomp, double ff[], void *userdata):
    # Convert the C array xx to a Python list for easier handling
    cdef list x_vals = [xx[i] for i in range(ndim)]
    # Retrieve the Python integrand function from userdata
    cdef object py_func = <object>userdata
    # Evaluate the Python function and store the result
    ff[0] = py_func(x_vals)

# Expose a Python-callable function to run the integration
def integrate(py_integrand_func):
    return tryCuhre(py_integrand, <void *>py_integrand_func)

Step 2: Configure the Setup Script #

Create a setup.py to compile the Cython module and link against your libcuhre.so:

from setuptools import setup, Extension
from Cython.Build import cythonize

# Define the Cython extension
ext = Extension(
    name="cuhre_wrapper",
    sources=["cuhre_wrapper.pyx"],
    include_dirs=["."],  # Path to your cuhre.h
    libraries=["cuhre"],  # Link against libcuhre.so
    library_dirs=["."],  # Path to your libcuhre.so
    runtime_library_dirs=["."]  # Ensure the shared library is found at runtime
)

setup(
    name="Cuhre Integration Wrapper",
    ext_modules=cythonize(ext)
)

Step 3: Compile and Test #

1. Compile the Cython module:

python setup.py build_ext --inplace

2. Make sure your system can find libcuhre.so at runtime. Either move it to a standard library path (like /usr/local/lib) or set the LD_LIBRARY_PATH environment variable:

export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:.

3. Test with a simple Python script:

import cuhre_wrapper

def my_2d_integrand(x):
    # Example: integrate x0 + x1 over [0,1]x[0,1] (expected result: 1.0)
    return x[0] + x[1]

result = cuhre_wrapper.integrate(my_2d_integrand)
print(f"Integral Result: {result}")

Troubleshooting Tips #

• Undefined reference errors: Double-check that libcuhre.so was compiled with -lcuba and that the Cuba library path is correct.
• Runtime library not found: Verify LD_LIBRARY_PATH includes the directory with libcuhre.so, or use rpath flags in the Extension definition to embed the path.
• Integrand type mismatches: Ensure your cuhre.h correctly includes cuba.h so integrand_t is properly defined—this is critical for matching the function signature expected by tryCuhre.

内容的提问来源于stack exchange，提问作者paulinho