The BookLeaf code is structured into 4 main categories: common, packages, utilities and infrastructure.
common contains general code handling data storage and data manipulation interfaces, general runtime configuration and error handling.
Each package is designed to expose a self-contained set of functionality. For example the hydro package contains the code implementing the hydrodynamics physics. Within packages, functionality is separated into drivers and kernels. Client code should only call the drivers, which will delegate to kernels to do the bulk of the work.
Package drivers can call utilities, which are similar to packages but contain functionality used by multiple packages. Utilities reduce code duplication. The distributed communication utility in comms is an example of a utility that is used by multiple packages.
Finally, the infrastructure code organises calls to package drivers and utilities to structure the simulation as a whole.
The solver() method in the infrastructure package contains the main loop. The
following diagram shows the structure of each iteration (coloured coded by
package and utility):
BookLeaf's main computational core consists of approximately 35 kernels which are combined to advance the simulation state each timestep. Here we use the term kernel to mean a self-contained loop over the mesh, which does some computation and updates one or more quantities. The kernels are designed to be suitable for efficient execution on modern parallel hardware. The rest of the code handles coordinating the execution of the kernels.