Error on Windows

[fbordeu]

Hi

I have an installation of python + mkl + pypardiso using conda-forge and I cant make pardiso to work on the example. This is a log to reproduce the problem.

Any ideas??? Thanks

C:\Users\XXXXXXX>mamba create -n PT python=3.9 mkl numpy scipy pypardiso

Looking for: ['python=3.9', 'mkl', 'numpy', 'scipy', 'pypardiso']

conda-forge/noarch 85.0MB @ 5.9MB/s 14.5s conda-forge/win-64 153.4MB @ 5.0MB/s 30.9s Transaction

Prefix: C:\Users\XXXXXXX\AppData\Local\miniforge3\envs\PT

Updating specs:

• python=3.9
• mkl
• numpy
• scipy

• pypardiso

  Package Version Build Channel Size

  Install:

  • python_abi 3.9 4_cp39 conda-forge Cached
  • ucrt 10.0.22621.0 h57928b3_0 conda-forge Cached
  • ca-certificates 2024.2.2 h56e8100_0 conda-forge Cached
  • intel-openmp 2024.0.0 h57928b3_49841 conda-forge Cached
  • vc14_runtime 14.38.33130 h82b7239_18 conda-forge Cached
  • vs2015_runtime 14.38.33130 hcb4865c_18 conda-forge Cached
  • vc 14.3 hcf57466_18 conda-forge Cached
  • libiconv 1.17 hcfcfb64_2 conda-forge Cached
  • pthreads-win32 2.9.1 hfa6e2cd_3 conda-forge Cached
  • tk 8.6.13 h5226925_1 conda-forge Cached
  • openssl 3.2.1 hcfcfb64_0 conda-forge Cached
  • libsqlite 3.45.1 hcfcfb64_0 conda-forge Cached
  • libzlib 1.2.13 hcfcfb64_5 conda-forge Cached
  • libffi 3.4.2 h8ffe710_5 conda-forge Cached
  • bzip2 1.0.8 hcfcfb64_5 conda-forge Cached
  • xz 5.2.6 h8d14728_0 conda-forge Cached
  • libxml2 2.12.5 hc3477c8_0 conda-forge Cached
  • libhwloc 2.9.3 default_haede6df_1009 conda-forge Cached
  • tbb 2021.11.0 h91493d7_1 conda-forge Cached
  • mkl 2024.0.0 h66d3029_49658 conda-forge Cached
  • tzdata 2024a h0c530f3_0 conda-forge Cached
  • llvm-meta 5.0.0 0 conda-forge Cached
  • python 3.9.18 h4de0772_1_cpython conda-forge Cached
  • openmp 5.0.0 vc14_1 conda-forge Cached
  • libflang 5.0.0 h6538335_20180525 conda-forge Cached
  • libopenblas 0.3.26 pthreads_hc140b1d_0 conda-forge Cached
  • libblas 3.9.0 21_win64_openblas conda-forge Cached
  • libcblas 3.9.0 21_win64_openblas conda-forge Cached
  • liblapack 3.9.0 21_win64_openblas conda-forge Cached
  • numpy 1.26.4 py39hddb5d58_0 conda-forge Cached
  • scipy 1.12.0 py39hddb5d58_2 conda-forge Cached
  • pypardiso 0.4.3 pyhd8ed1ab_0 conda-forge Cached

  Summary:

  Install: 32 packages

  Total download: 0 B

---

Confirm changes: [Y/n] Y Downloading and Extracting Packages:

Preparing transaction: done Verifying transaction: done Executing transaction: done

To activate this environment, use

 $ mamba activate PT

To deactivate an active environment, use

 $ mamba deactivate

C:\Users\XXXXXXX>mamba activate PT

C:\Users\XXXXXXX>python Python 3.9.18 | packaged by conda-forge | (main, Dec 23 2023, 16:29:04) [MSC v.1929 64 bit (AMD64)] on win32 Type "help", "copyright", "credits" or "license" for more information.

import pypardiso import numpy as np import scipy.sparse as sp A = sp.rand(10, 10, density=0.5, format='csr') A <10x10 sparse matrix of type '<class 'numpy.float64'>' with 50 stored elements in Compressed Sparse Row format> b = np.random.rand(10) x = pypardiso.spsolve(A, b) Traceback (most recent call last): File "", line 1, in File "C:\Users\XXXXXXX\AppData\Local\miniforge3\envs\PT\lib\site-packages\pypardiso\scipy_aliases.py", line 46, in spsolve solver.factorize(A) File "C:\Users\XXXXXXX\AppData\Local\miniforge3\envs\PT\lib\site-packages\pypardiso\pardiso_wrapper.py", line 164, in factorize self._call_pardiso(A, b) File "C:\Users\XXXXXXX\AppData\Local\miniforge3\envs\PT\lib\site-packages\pypardiso\pardiso_wrapper.py", line 277, in _call_pardiso self._mkl_pardiso(self.pt.ctypes.data_as(ctypes.POINTER(self._pt_type[0])), # pt OSError: [WinError -1066598273] Windows Error 0xc06d007f

[ghsilva]

Long story short, I believe using MKL version 2023.2.0 or older should solve this problem while we are working in a patch release for the intel-openmp 2024.0.*

[fbordeu]

Thanks for the hint,

I'm going to try lowering the version of intel-openmp. So this problem also exists on others OS(Linux)?

Thanks for the help

[ghsilva]

Note that MKL 2024. requires Intel-openmp 2024. so you will need to lower the MKL version as well. The issue happens only on Windows.

[fbordeu]

Exactly, mamba did the correct job of keeping my env functional by lowering ihe intel-openmp and the mkl.

Thanks,

[haasad]

I have now also released pypardiso v0.4.4 on conda-forge, which pins MKL to <2024, see https://github.com/conda-forge/pypardiso-feedstock/pull/9