PETSc

PETSc is the Portable, Extensible Toolkit for Scientific Computation.

Version 3.1-p8

(March 25, 2010)

That's the version needed by belt. When choosing to download packages during configuration, versions are:

• openmpi-1.4.1.tar.gz
• mpich2-1.0.8.tar.gz
• hypre-2.6.0b.tar.gz (uses deprecated MPI_Address(), MPI_Type_struct)

Problem when using system's MPI (more modern):

• PETSc 3.1-p8 uses deprecated objects: MPI_Attr_get, MPI_Attr_put, MPI_Attr_delete, MPI_Keyval_create, MPI_Keyval_free, OMPI_C_MPI_NULL_COPY_FN, OMPI_C_MPI_NULL_DELETE_FN (deprecated since MPI-2.0) and MPI_Errhandler_create, MPI_Errhandler_set, MPI_Type_struct (removed in MPI-3.0)
• Pluto4.1 itself uses deprecated objects (in Parallel/al_subarray_.c): MPI_Type_extent, MPI_Type_hvector, MPI_Type_struct (removed in MPI-3.0)

OpenMPI is (at least on Debian) hardwired to reject removed functions (#define OMPI_ENABLE_MPI1_COMPAT 0 in /usr/lib/x86_64-linux-gnu/openmpi/include/mpi.h and /usr/lib/x86_64-linux-gnu/fortran/gfortran-mod-15/openmpi/mpi.h). Changing 0 → 1 (by the admin) works, the compiler flag -Wno-deprecated-declarations is recommended, then. A script like the following for configuring PETSc is advisable:

Debian (merope, asterope, electra)

#!/bin/bash

# L.B. 2024

declare -A inc=(["openmpi"]="/usr/lib/x86_64-linux-gnu/openmpi/include",
                ["mpich"]="/usr/include/x86_64-linux-gnu/mpich")

declare -A lib=(["openmpi"]="[/usr/lib/x86_64-linux-gnu/libmpi.so,/usr/lib/x86_64-linux-gnu/libmpi_mpifh.so]",
                ["mpich"]="[/usr/lib/x86_64-linux-gnu/libmpich.so,/usr/lib/x86_64-linux-gnu/libmpichfort.so]")

# On Debian 12, the MPICH libraries are actually called
# "libmpich.so.12" and "libmpichfort.so.12", and the standard links
# .so -> .so.12 are missing. Ask your admin to fix that. Specifying
# the suffix ".12" doesn't work, it confuses the configure.py.

export PETSC_DIR=$PWD

opt=-O3  # Some prefer -O2.
mpi=openmpi

python2 ./config/configure.py PETSC_ARCH=debian_$mpi \
        --with-cc=mpicc.$mpi --with-cxx=mpicxx.$mpi --with-fc=mpif90.$mpi --with-mpiexec=mpirun.$mpi \
        --CFLAGS=-Wno-deprecated-declarations --COPTFLAGS=$opt --FOPTFLAGS=$opt \
        --with-x=0 --with-debugging=0 --download-hypre=1 \
        --with-mpi-include="${inc[$mpi]}" --with-mpi-lib="${lib[$mpi]}"

magnitude

#!/bin/bash

# L.B. 2024

module -s load compiler/latest
module -s load mkl/latest
module -s load mpi/latest

export PETSC_DIR=$PWD

python2 ./config/configure.py PETSC_ARCH=magnitude_intel \
        --CFLAGS= --COPTFLAGS=-O3 --FOPTFLAGS=-O3 \
	    --with-blas-lib=[libmkl_intel_lp64.a,libmkl_sequential.a,libmkl_core.a] --with-lapack-lib=libmkl_core.a \
        --with-x=0 --with-debugging=0 --download-hypre=1

# The Intel® oneAPI Math Kernel Library ILP64 libraries use the 64-bit
# integer type (necessary for indexing large arrays, with more than
# 2^31-1 elements), whereas the LP64 libraries index arrays with the
# 32-bit integer type.

In your script or shell, where you do the actual make (as given in the last line of output of the configure script), you need first to load above modules, too.