如何查看Python链接的二进制库文件？含DyNet安装验证

Let’s walk through concrete steps to confirm your manual MKL-enabled DyNet build is the one Python is using:

• Check DyNet’s build configuration directly in Python
  Fire up a Python shell and run these commands:

  import dynet as dy
  print(dy.__version__)
  print(dy.config())
  

  The dy.config() output will list all compile-time flags. Look for a line like USE_MKL: ON—this confirms MKL support was enabled during your recompile.

• Verify the module’s file path matches your manual build
  To ensure Python isn’t still loading the old pip-installed version, check where DyNet is being pulled from:

  import dynet as dy
  print(dy.__file__)
  

  This should point to the .so (Linux/macOS) or .pyd (Windows) file in the directory where you manually compiled and installed DyNet—not the default site-packages location from your initial pip install.

• Optional: Quick performance sanity check
  MKL optimizes linear algebra operations, so a simple test can hint at whether it’s active. Run something like:

  import dynet as dy
  import time
  
  dy.renew_cg()
  mat = dy.random_normal((1000, 1000))
  start = time.time()
  mat * mat
  dy.forward()
  print(f"Matrix multiplication took {time.time() - start:.2f} seconds")
  

  Compare this to the speed before your MKL rebuild—you should see a noticeable speedup if MKL is working (results vary by hardware, but large tensor operations will feel significantly faster).

For any library that wraps C++ code and exposes a Python API, here are reliable ways to dig into its details:

• Locate the underlying binary file
  Almost all Python modules (including C extensions) expose a __file__ attribute pointing to their on-disk location. For C++-backed libraries, this is the compiled shared object file:

  import your_cpp_library
  print(your_cpp_library.__file__)
  

  If __file__ fails, use the inspect module as a fallback:

  import inspect
  import your_cpp_library
  print(inspect.getfile(your_cpp_library))
  

• Check build/version metadata
  Many libraries include helper methods to output compile-time configurations or version details. Examples include:

  • OpenCV: cv2.getBuildInformation()
  • PyTorch: torch.__config__.show()
  • DyNet (as we used): dy.config()
    These outputs will tell you about enabled features, linked dependencies, and compiler settings.

• Inspect system-level dependencies
  To see exactly which system libraries the C++ extension links to (like MKL), use platform-specific tools:

  • Linux: Run ldd /path/to/your/library.so in the terminal—look for MKL-related libraries like libmkl_rt.so or libmkl_core.so.
  • macOS: Use otool -L /path/to/your/library.so to list linked libraries.
  • Windows: Use dumpbin /dependents C:\path\to\your\library.pyd (requires Visual Studio tools in your PATH).

内容的提问来源于stack exchange，提问作者Erel Segal-Halevi