Parallelisation

The flesh itself does not actually set up grid variables. This is done by a driver thorn. To allow the distribution of a grid over a number of processors, the driver thorn must also provide the grid decomposition, and routines to enable parallelisation. The method used to provide this parallelisation (e.g. MPI, PVM) is not usually important for the thorn writer, since the driver thorn provides routines which are called by standard interfaces from the flesh. Here, we describe briefly the most important of these routines for the application thorn writer. A more detailed description of these interfaces with their arguments, is given in the Reference Manual. A complete description of the routines that a driver thorn must provide, will be provided in the Infrastructure Thorn Writers Guide (Part D). The standard driver thorn is currently PUGH in the CactusPUGH package, which is a parallel unigrid driver.

CCTK_nProcs

Returns the number of processors being used

CCTK_MyProc

Returns the processor number (this starts at processor number zero)

CCTK_SyncGroup, CCTK_SyncGroupsI

Synchronises either a single group or a set of groups of grid arrays by exchanging the values held in each processor ghostzones, with the physical values of their neighbours (see the Reference Manual)

CCTK_Barrier

Waits for all processors to reach this point before proceeding