The idea of this library was inspired a bit by a very cool talk Chris Lattner gave
at some conference. In his talk, he mentions how the LLVM project just grew in
many directions thanks to how the design of the project was done. I.e. the fact
that most of the components of the LLVM are in the lib subdirectory.
If you’ve never compiled the LLVM I suggest you do it, you’ll see that you can enable different compoenents depending on your needs. Rarely does one need to build everything in it.
The same idea will be applied to PIC. I want to provide a set of components
that can be used independently of each other. There will be of course, a core library
that will provide the most basic functionality. For example, everything at the very bottom depends
on the pic_types.F90 module, so you cannot really use PIC without it.
So following this concept, PIC will create a series of libraries that are based upon the core library, i.e. the very basic functionality that will always be built no matter what project you select on top.
This idea is more important if you’re using the CMake build system, since this
will need to be configured when building from source. Otherwise, if you are using
the Fortrn package manager and using pic as a dependency in your project, the FPM
will only builds the modules/components that you actually include in your project.
I find this very neat.
At this point, the core library is everything, I haven’t refactored the code to be built into separate libraries, however I already have the modularity to not build certain bits, like BLAS, or MPI. I envision the following structure
libpic_core
libpic_blas
libpic_mpi
libpic_gpu
And PIC will be able to build all of these with default_int set to either int32 or int64, depending on the needs of the user.
Naturally, the MPI module will only use int32 as the default integer type, since MPI (mpi_f08) does not support int64 in its API.
I hope that this library will be useful for many people, mostly as a fast and efficient way to write parallel code in Fortran. I also hope that it will be a good example of how to write (modern) Fortran code that is modular, extensible, and easy to use.
This is also my attempt into showcasing that Fortran is not a boring old language that only has a relevance in legacy code.
I want to show that Fortran can be used to write modern, high-performance code that is easy to read and maintain.