MUMPS sparse solver

CMake downloads the unmodified source tarfile from MUMPS developers and builds. For offline usage, the user can specify a local MUMPS source archive like cmake -Bbuild -DMUMPS_url=/path/to/mumps.zip

CMake builds MUMPS in parallel faster and more conveniently than the original Makefiles. CMake allows easy reuse of MUMPS in external projects via any of:

• CMake FetchContent
• build MUMPS, cmake --install, then find_package(MUMPS CONFIG REQUIRED)
• CMake ExternalProject_Add (similar to FetchContent example)

MUMPS CeCILL-C license is distinct from this CMake script license. MUMPS teams typically make new releases each year.

Many compilers and systems are supported by CMake build system on Windows, MacOS and Linux. Static (default) or Shared cmake -DBUILD_SHARED_LIBS=on MUMPS builds are supported.

Virtually all contemporary compilers work, including GCC, Clang/Flang, oneAPI, NVHPC, AOCC, Cray, etc.

By default PORD ordering is used. Scotch, METIS, and parMETIS ordering can be used and will be automatically built if needed.

Several LAPACK vendors are supported.

The MUMPS project is distinct from this CMake script wrapper. See the MUMPS Users email list and MUMPS User Guide for any questions about MUMPS itself.

Build

From this repo's top directory:

sh cmake -B build cmake --build build

Numerous MUMPS build options are available.

The default precision is float64 and float32. To build all precisions including complex64, complex128 do:

```sh cmake -Bbuild -DBUILDSINGLE=on -DBUILDDOUBLE=on -DBUILDCOMPLEX=on -DBUILDCOMPLEX16=on

cmake --build build ```

The headers and library are optionally installed to an install prefix:

```sh cmake -Bbuild --install-prefix /path/to/install/mumps

cmake --install build ```

With the default options the build/ directory contains library binaries (Windows binaries have different names):

• libdmumps.a (real64)
• libsmumps.a (real32)
• libmumps_common.a (common MUMPS routines)
• libpord.a (PORD library)

If -DMUMPS_parallel=no was set, additional helper libraries are built in place of linking MPI libraries:

• libmpiseq_fortran.a
• libmpiseq_c.a

The libmpiseq.a file isn't used directly by this project, but it for compatibility with other build systems that may expect a libmpiseq file.

These libraries can be linked into C, C++, Fortran, etc. programs, or even be used with appropriate interfaces from Matlab and Python PyMUMPS and python-mumps.

Self test and examples

Optionally, run self-tests:

sh ctest --test-dir build

To build the example, first "install" the MUMPS package-the default install location is under the MUMPS build/local directory:

```sh cmake --install build

cmake -S example -B example/build -DMUMPS_ROOT=build/local

cmake --build example/build ```

Using binary libraries

Linking the MUMPS binaries into a user-program is project-dependent. An example using the examples in this project with GNU GCC, using the "mpicxx" MPI compiler wrapper:

sh mpicxx ./example/d_example.cpp -I./build/local/include -L./build/local/lib -ldmumps -lmumps_common -lpord -lscalapack -lblacs -llapack -lblas -lgfortran

If -DMUMPS_parallel=no was used to build MUMPS, instead do:

sh g++ ./example/d_example.cpp -I./build/local/include -L./build/local/lib -ldmumps -lmumps_common -lpord -llapack -lblas -lmpiseq_fortran -lmpiseq_c -lgfortran