Confining imperative languages for parallel processing

摘要

Imperative languages have built-in notions of global updatablevariables. This has been claimed to be one of the major reasons why theyare inferior to functional languages for parallel processing. The paperexamines the `bilateral' effects-side effects and aliasing-of updatablevariables. While aliasing hinders code parallelization, side effects donot necessarily cause problems in parallel processing if a certaindiscipline, e.g. a high-level dataflow scheme, known as the LGDGcomputation, in their use is obeyed. Furthermore, a point is made forthe exclusion of nested procedure declarations from parallel imperativelanguages, since they may cause serious aliasing and degrade locality ofreference