Enhancing the AKL compiler using global analysis

A method of enhancing the AKL platform, chiefly the compiler, using information provided by program analysis is presented. The analyzer is assumed to deliver information about the possible types that program variables can take, the possible aliases that the variable might have, and for variables that are bound to complex terms whether or not the term might have other references. The enhanced AKL compiler produces more optimal code, which when executed will improve execution performance, and save memory. On an abstract level it is shown that in favourable cases an invoked agent can more quickly reach the conclusion that it should suspend, suspend on fewer variables and thus create fewer suspensions in memory, and wake only to suspend again fewer times. Also, unification may be replaced by destructive assignment, saving memory, and even processing power if a structure that is being created has common fields with one that is being consumed. The report ends with a discussion about what can be expected from the optimization on real programs.