Selalib is the Semi Lagrangian Library, a collection of types and its associated methods that are useful for creating parallel plasma physics simulations that use the semilagrangian methodology for the solution of the Vlasov equation. In its design, we have attempted to expose to the users an interface that expresses as naturally as possible the problems that arise when using the semilagrangian approach. However, most of the modules contained in the library are not intrinsically related with the semilagrangian method, but are more general utilities which could be used in different contexts. A principle in the development of the library has been to continually identify specific functionalities and to attempt to promote them into their own standalone module, with a sufficiently general interface which may still permit a reasonably efficient implementation.
Selalib is structured in (loosely defined) layers. One can think about these layers as libraries in their own right. A given layer can use the capabilities offered by a lower layer through the exposed interface, but never from a higher level. The layer with the lowest level of abstraction presently contains basic utilities like memory allocators, assertions and basic numeric types. The second layer is composed of numerical and parallel utilities. The third and highest level of the library contains the semi-lagrangian methodology tools, types and methods. This manual will ultimately describe each of these layers and the modules therein.
Most of the native types and operations provided by Selalib are prefixed
by sll_. In this way you can at least have an expectation of finding
documentation (if a user) and a starting point of where to start looking
if you wish to dive into some particular aspect of an implementation (if
a developer). This convention has not been always followed, so it would
be advantageous to slowly but surely include this prefix throughout as a
first means to create a Selalib namespace.
Selalib is written in Fortran 2003. We have deviated from the original intent to make Selalib fully f95-compatible for multiple reasons: the newer standard permits a separation of interface and implementation which is desirable in a development like this, the reduction in coding that is obtained with the use of abstract classes is appealing for a project with quite limited manpower as this one, and finally, the desire to make this product as forward-looking as possible.
Some features of the library are implemented as macros. To the user of
the library, it makes no difference whether some functionality is
implemented in the form of a procedure or a macro, with the exception
that presently, macro names are written in ALL-CAPS. The use of the
macro is required to offer certain capabilities, like informative error
messages. For a developer, the use of the macros is needed in many cases
to reduce code redundancy. The need for the use of macros will hopefully
be more understandable when the reader sees the behavior of calls to
simple macros like SLL_ALLOCATE() or SLL_ASSERT().
A macro is a pre-processor directive. Selalib uses only very simple
macros that are handled by fpp, the Fortran Pre-Processor. fpp's
capabilities are very limited and one peculiarity of its output is that
macros are expanded into a single long line, which can easily surpass
the 132 character limit that Fortran systems have. For this reason
alone, the compilation of Selalib requires the use of a compilation
flag: -ffree-line-length-none (in gfortran) or its equivalent in
another compiler. On a similar vein, some compilers require the
extension .F90 (as opposed to .f90) in order to apply a
preprocessing step. Thus, all files in Selalib use the .F90 extension.
Clients of the library which want to use the macro facilities should
thus also use this extension as well.