A SIMPLE DATA DEPENDENCY ANALYZER FOR C PROGRAMS

摘要

Data dependencies that exist in a sequential program are a major hindrance towards parallelization. Dependencies between simple statements are relatively easy to detect. However, there are a lot of dependencies that are hidden due to procedure calls. Most of interprocedural data dependencies are caused by the use of global variables, passing of information via parameters and also by aliasing. Furthermore, data dependency analysis is a complex task, requiring much analysis and resources. Our work reports on the development of a simple interprocedural data dependency analyzer for C programs incorporating simple techniques. A read and write set is created for each statement, 'which are then compared for overlapping variable names. Using this technique, we are able to check dependencies between statements, across procedures, within a procedure, array loops, 1F-THEN-ELSE blocks as well as standard library function calls. Results from this analyzer is a dependency table which can be used by a parallelizing compiler.