Recipe and Binaries of MUMPS for MATLAB with OpenMP Support

This repository contains the recipe for building multithreaded MUMPS for MATLAB on Linux, Mac, and Windows. The multithreading relies on the OpenMP version of OpenBLAS as well as some additional OpenMP features in MUMPS.

Binary Distributions of MEX files

The easiest way for MATLAB users is to download the MEX files prebuilt for your system.

	double precision real	double precision complex
Linux(1)	dmumpsmex.mexa64	zmumpsmex.mexa64
Windows(2)	dmumpsmex.mexw64	zmumpsmex.mexw64
Mac(3)	dmumpsmex.mexmaci64	zmumpsmex.mexmaci64

These MEX files were built with MUMPS v5.3.4, OpenBLAS v0.3.21, METIS 5.1.0, and MATLAB R2022a. They work with MUMPS's MATLAB interface, which is included in the MATLAB directory in this repository. Simply download these prebuilt MEX files in the MATLAB folder.

Notes:

1. The MEX files should work out of the box on Linux systems in both desktop and command-line mode. However, when running in command-line mode, you should start matlab with -nojvm option instead of simply -nodesktop or -nodisplay, or JAVA is prone to throwing errors during the shutdown process.
2. For Windows users, you must also install MinGW-w64 C/C++ compiler, which comes with provides libgfortran and libgomp required by OpenBLAS and MUMPS.
3. For Mac users, you need to create a soft link /Applications/MATLAB_<MATLAB_VERSION>.app/sys/os/maci64/libomp.dylib pointing to libiomp5.dylib in the same directory. In addition, you need to add three gfortran libraries to MATLAB's sys/os/maci64 folder, namely libgfortran.5.dylib, libquadmath.0.dylib, libgcc_s.1.1.dylib. If you already installed gfortran using anaconda, miniconda, or homebrew, you could create soft links to these files in the /Applications/MATLAB_R2022a.app/sys/os/maci64 folder. Otherwise, you can simply run the following command:

   curl -Ls https://github.com/xmjiao/mumps4m-openmp/releases/download/v5.3.4/libgfortran_dylibs.tgz | tar zxv - -C /Applications/MATLAB_R2022a.app/sys/os/maci64

Of course, you need to replace R2022a with the version of your MATLAB installation.

Binary Distributions of Static Libraries

If you want to link MUMPS into your own MEX files for MATLAB, for example, to support single-precision arithmetic, you can download the following prebuilt static libraries of MUMPS, OpenBLAS, and METIS. See https://github.com/xmjiao/mumps4m-openmp/tree/main/recipe/MATLAB to see how these files can be used in the mex command.

	Static libraries
Linux	libmumps_5.3.4_Linux-x86_64.tar.gz
Windows	libmumps_5.3.4_MINGW64-x86_64.tar.gz
Mac	libmumps_5.3.4_Darwin-x86_64.tar.gz

AUTHOR

mumps4m-openmp was developed and maintained by Xiangmin (Jim) Jiao (xiangmin.jiao@stonybrook.edu) mainly for comparative research. If you need robust and efficient linear solvers for large-scale problems, please also consider the software hifir. For comparisons between HIFIR and MUMPS, please see the following papers:

@Article{chen2021hilucsi,
  author  = {Chen, Qiao and Ghai, Aditi and Jiao, Xiangmin},
  title   = {{HILUCSI}: Simple, robust, and fast multilevel {ILU} for
             large-scale saddle-point problems from {PDE}s},
  journal = {Numer. Linear Algebra Appl.},
  year    = {2021},
  number  = {6},
  pages   = {e2400},
  volume  = {28},
  doi     = {10.1002/nla.2400}
}

@Article{chen2022hifir,
  author  = {Chen, Qiao and Jiao, Xiangmin},
  title   = {{HIFIR}: Hybrid incomplete factorization with iterative refinement
             for preconditioning ill-conditioned and singular systems},
  journal = {ACM Trans. Math. Softw.},
  year    = {2022},
  doi     = {10.1145/3536165}
}