8

I'm trying to create an executable python program that runs on windows without python being installed, for this I'm using cx_Freeze. But I get the following error: "Cannot load mkl_intel_thread.dll"

On my PC, which has python installed (miniconda3), I built the executable using cx_Freeze, and when I ran the executable I also would get "Cannot load mkl_intel_thread.dll". I fixed this by going to my python folder, Library\bin, and copied the mkl_intel_thread.dll file to where the executable is placed. The problem is, when moving the whole folder to another PC (without python installed), this error reappears, even though the mkl_intel_thread.dll is in the folder.

File that I want to distribute (plot.py):

import matplotlib.pyplot as plt

a = [0, 1, 2]
b = [0, 2, 0]
plt.fill(a, b, 'b')
plt.show()

cx_Freeze setup file (setup.py):

import cx_Freeze
import sys
import matplotlib
import numpy
import os

os.environ['TCL_LIBRARY'] = "C:\\Miniconda3\\tcl\\tcl8.6"
os.environ['TK_LIBRARY'] = "C:\\Miniconda3\\tcl\\tk8.6"


executables = [cx_Freeze.Executable("plot.py")]


build_exe_options = {"includes":['numpy.core._methods',
        'numpy.lib.format', 'matplotlib.backends.backend_tkagg']}

cx_Freeze.setup(
    name = "script",
    options = {"build_exe": build_exe_options},
    version = "0.0",
    description = "A basic example",
    executables = executables)
Carlos
  • 586
  • 1
  • 9
  • 19

4 Answers4

9

EDIT:

  1. Try to copy all files starting with mkl you find under Library\bin or numpy\core into the build folder, as well as libiomp5md.dll, see Python Pyinstaller 3.1 Intel MKL FATAL ERROR: Cannot load mkl_intel_thread.dll and cx_freeze converted GUI-app (tkinter) crashes after pressing plot-Button.

    Once you have found out which file(s) need(s) to be manually copied, you can let cx_Freeze include the necessary file(s) by using the include_files list of the build_exe options (see code snippet below). If necessary, you can use a tuple (source, destination) as item in the include_files list to let cx_Freeze copy a file from source to a specific destination into the build directory, see the cx_Freezedocumentation.

  2. I see further potential problems in the setup script you've posted in your question:

    • include the whole numpy packages using the packages list of the build_exe options, it is easier and maybe safer
    • it is safer to dynamically find out the location of the TCL/TK DLLs
    • for cx_Freeze 5.1.1, the TCL/TK DLLs need to be included in a lib subdirectory of the build directory

In summary, try t o use

PYTHON_INSTALL_DIR = os.path.dirname(os.path.dirname(os.__file__))
os.environ['TCL_LIBRARY'] = os.path.join(PYTHON_INSTALL_DIR, 'tcl', 'tcl8.6')
os.environ['TK_LIBRARY'] = os.path.join(PYTHON_INSTALL_DIR, 'tcl', 'tk8.6')

build_exe_options = {'packages': ['numpy'],
                     'includes': ['matplotlib.backends.backend_tkagg'],
                     'include_files': [(os.path.join(PYTHON_INSTALL_DIR, 'DLLs', 'tcl86t.dll'),
                                        os.path.join('lib', 'tcl86t.dll')),
                                       (os.path.join(PYTHON_INSTALL_DIR, 'DLLs', 'tk86t.dll'),
                                        os.path.join('lib', 'tk86t.dll'))
                                       # add here further files which need to be included as described in 1.
                                      ]}

in your setup script.

jpeg
  • 2,372
  • 4
  • 18
  • 31
  • on my pc this will still give me "Intel MKL FATAL ERROR: Cannot load mkl_intel_thread.dll.". But again, I can copy the dll to the folder and it will work.I'll transfer to the other pc later but I suspect it will give the same error. – Carlos Jan 25 '19 at 17:06
  • 2
    Try to copy all files starting with `mkl` you find under `Library\bin` or `numpy\core` into the build folder, as well as `libiomp5md.dll`, see [this post](https://stackoverflow.com/q/35078548/8516269) and [this post](https://stackoverflow.com/a/34893933/8516269). – jpeg Jan 28 '19 at 06:53
  • @Carlos Sounds great. I've edited my answer accordingly. – jpeg Jan 30 '19 at 07:51
1

A similar issue affects cx_Freeze 6.1 or 6.2: the executable does not launch, either without error message or with

INTEL MKL ERROR: The specified module could not be found. mkl_intel_thread.dll.
Intel MKL FATAL ERROR: Cannot load mkl_intel_thread.dll.

Configuration:

This is also observed with Python 3.6.8 or earlier versions of numpy such as e.g. 1.18.4+mkl or 1.19.0+mkl.

I've observed that cx_Freeze includes 3 DLLs mkl_rt.dll, python38.dll, and vcruntime140.dll in the subdirectory lib\numpy\core of the build directory, whereas the original installation does not contain any DLL in the subdirectory site-packages\numpy\core (all DLLs are in site-packages\numpy\DLLs). If I manually remove mkl_rt.dll from the subdirectory lib\numpy\core of the build directory after building the application with cx_Freeze, the issue disappears and the application works.

This solution can be implemented by adding the following code at the end of the setup.py script:

numpy_core_dir = os.path.join(dist_dir, 'lib', 'numpy', 'core')
for file_name in os.listdir(numpy_core_dir):
    if file_name.lower().endswith('.dll'):
        file_path = os.path.join(numpy_core_dir, file_name)
        os.remove(file_path)

where dist_dir is the build directory generated by cx_Freeze (passed to the build_exe option).

jpeg
  • 2,372
  • 4
  • 18
  • 31
0

Just copy these four files in cx_freeze generated build folder

mkl_core.dll
mkl_def.dll
mkl_intel_thread.dll
mkl_mc3.dll
David Buck
  • 3,752
  • 35
  • 31
  • 35
sultan
  • 1
0

Manage to find a solution to this by downgrading numpy==1.18.2 from numpy==1.19.1 when using cx_Freeze==6.5.3.

Ivan
  • 311
  • 3
  • 7