PSyclone Transformation Scripts

The LFRic build system searches for PSyclone transformation scripts prior to applying any PSyclone transformations. Initially, the build system will search for a Python (.py) module with a matching filename to the Fortran module currently being operated on. If a match is made, the build system proceeds to call the psyclone command on the current module, specifying the path to the transformation script in the command. If a match is not made, the build system proceeds to search for a more generic transformation script, global.py, and applies it to the current module via PSyclone. The transformations called in global.py will be applied to any module that does not have a transformation script of its own. This principle is true for both LFRic and non-LFRic source. Finally, if no optimisation script is present, global or otherwise, the module is passed to PSyclone with no transformation script specification. This will also remove any existing directives in the source file.

Note

LFRic is written in the PSyKAl format and therefore uses PSyclone’s psykal method of operation to preprocess modules. There is no support for other PSyclone methods of operation in the LFRic Core library.

To direct the build system to find a transformation script, certain GNU Make variables must be set. Namely, the OPTIMISATION_PATH and DSL variables.

• The OPTIMISATION_PATH variable should be set to the common path between all transformation scripts within an application.

• The DSL variable directs the build to search in specified sub-directories of the OPTIMISATION_PATH.

The purpose of DSL is to separate transformations that would otherwise conflict. For example, in the skeleton miniapp, transformation scripts are grouped by the PSyclone method of operation they require (psykal, transmute):

optimisation/
└── platform/ (ex1a, minimum, xc40, etc.)
    ├── psykal
    │   ├── global.py
    │   └── sub_directory/
    └── transmute/
        ├── global.py
        └── sub_directory/