The algorithms of automatic mapping of Fortran programs to cluster are considered. The algorithms use the results
of the program static analysis, which, if necessary, can be corrected or supplemented by a programmer. Automatic
mapping to cluster is based on the construction of various parallelization variants (data distribution and computation
distribution between processors, as well as providing access to remote data on other processors) and their
evaluation by modeling the parallel execution of their respective programs.
|
