Incremental dependence analysis
Rosene, Carl M.
Doctor of Philosophy
New supercomputers depend upon parallel architectures to achieve their high rate of computation. In order to take advantage of the power of such a machine, programs must be executed in parallel, that is, parts of the program must execute on different processors. When a program is executed in parallel it is impossible to guarantee the execution order of the parts of the program being executed by different processors. Dependence analysis identifies the statements whose execution order must be preserved in order for the program to be correct. Making available the results of dependence analysis in an interactive programming environment can aid a programmer in writing programs which will execute more efficiently on a parallel machine. In order to provide the results of dependence analysis to the programmer in as timely a manner as possible, incremental methods of dependence analysis have been developed. The performance of these algorithms has been estimated based on static measurements of F scORTRAN programs. This dissertation presents the incremental methods and the results of estimates of their performance.