Optional third-party tools and libraries

Table of Contents

• OpenBLAS / libxsmmJIT / Lapack
• HDF5
• The Visualization Toolkit (VTK)

• Otter task tracing

OpenBLAS / libxsmmJIT / Lapack

OpenBLAS is an implementation of BLAS, which provide some basic linear algebra subroutines. libxsmmJIT (available as submodule) and lapack (more complicated linear algebra routines, used for matrix inversion etc) depend on this.

We provide a few configure options to make sure all the headers and libraries are available. When running configure script (see Installation) you should include

-openblas

to enable OpenBLAS. Further, you should use with-lapack to enable lapack and similarly with-libxsmmjit. Configuration for these two options will fail without OpenBLAS enabled.

We leave it up to the user to modify the CXXFLAGS, LDFLAGS and LD_LIBRARY_PATH correctly. These options were tested on COSMA (see Reference configurations for some machines), where installations of openblas were already available.

When configuring, be sure to add:

• A link to the include directory for the version of openblas for your compiler to CXXFLAGS
• A link to the lib directory for the version of openblas for your compiler to LDFLAGS
• A link to the lib directory for the version of openblas for your compiler to LD_LIBRARY_PATH
• -lopenblas to LDFLAGS

The call to configure should look something like this (for example, on COSMA):

```./configure –with-openblas –with-lapack [other configure options] CXXFLAGS="... -I/cosma/local/openblas/intel_2020-update2/0.3.12cosma7/include/" LDFLAGS="-L/cosma/local/openblas/intel_2020-update2/0.3.12cosma7/lib/ -lopenblas" ...

Make sure to export your LD_LIBRARY_PATH first.

HDF5

Peano does not need hdf5. There's no generic input file format, and it can write its own mesh files without hdf5. Things are different if you want to dump hdf5, or if you use an extension/toolbox which requires hdf5. In this case, you have to enable hdf5 in the configuration (pass in –with-hdf5) and ensure that your CXXFLAGS and LDFLAGS include all libraries.

For the latter, different systems have other paths. On local Ubuntu workstations, we typically use something similar to

-L/usr/lib/x86_64-linux-gnu/hdf5/serial -lhdf5 -lhdf5_cpp

Please note that two hdf5 libraries are required: the hdf5 library which is written in C and the C++ counterpart.

The Visualization Toolkit (VTK)

Peano usually dumps data data in its own native file format. However, there are several extensions where it directly writes into VTK files or cousins of it, such that you can directly visualise the outcome with Paraview of VisIt, e.g. Most users do not need to build Peano directly against any third-party visualisation libraries.

However, ...