Awesome

GridapPETSc

GridapPETSc is a plugin of GridapDistributed.jl that provides the full set of scalable linear and nonlinear solvers in the PETSc library. It also provides serial solvers to Gridap.jl.

Documentation

Take a look at this tutorial for learning how to use GridapPETSc in distributed-memory simulations of PDEs.

It can also be used in the serial case, as shown in this test.

Installation

GridapPETSc julia package requires the PETSC library (Portable, Extensible Toolkit for Scientific Computation) and MPI to work correctly. You have two main options to install these dependencies.

• Do nothing [recommended in most cases]. Use the default precompiled MPI installation provided by MPI.jl and the pre-compiled PETSc library provided by PETSc_jll. This will happen under the hood when you install GridapPETSc. In the case of GridapPETSc, you can also force the installation of these default dependencies by setting the environment variable JULIA_PETSC_LIBRARY to an empty value.

• Choose a specific installation of MPI and PETSc available in the system [recommended in HPC clusters].

  • First, choose a MPI installation. See the documentation of MPI.jl for further details.
  • Second, choose a PETSc installation. To this end, create an environment variable JULIA_PETSC_LIBRARY containing the path to the dynamic library object of the PETSC installation (i.e., the .so file in linux systems). Very important: The chosen PETSc library needs to be configured with the MPI installation considered in the previous step.

Notes

• GridapPETSc default sparse matrix format is 0-based compressed sparse row. This type of sparse matrix storage format can be described by the SparseMatrixCSR{0,PetscReal,PetscInt} and SymSparseMatrixCSR{0,PetscReal,PetscInt} Julia types as implemented in the SparseMatricesCSR Julia package.
• When running in MPI parallel mode (i.e., with a MPI communicator different from MPI.COMM_SELF), GridapPETSc implements a sort of limited garbage collector in order to automatically deallocate PETSc objects. This garbage collector can be manually triggered by a call to the function GridapPETSc.gridap_petsc_gc(). GridapPETSc automatically calls this function inside at different strategic points, and this will be sufficient for most applications. However, for some applications, with a very frequent allocation of PETSc objects, it might be needed to call this function from application code. This need will be signaled by PETSc via the following internal message error PETSC ERROR: No more room in array, limit 256 recompile src/sys/objects/destroy.c with larger value for MAXREGDESOBJS