Data Flow Analysis across Tuplespace Process Boundaries

The increasing attention toward distributed shared memory systems attests to the fact that programmers nd shared memory parallel programming easier than message passing programming, while physically distributed memory multiprocessors and networks of workstations o er the desirable scalability for large applications. A current limitation of compilers for shared memory parallel languages is their restricted use of traditional scalar code-improving transformations, such as constant propagation and dead code elimination. The major problem lies in the failure of data ow analysis techniques developed for sequential programs in the context of shared memory programs with user-speci ed parallelism. Notable e orts to develop data ow frameworks for optimizing parallel programs have focused on programs with lexically-speci ed parallel constructs, such as cobegin/coend, where sections of the parallel constructs are data independent except where an appropriate synchronization mechanism is used. Not only do the data ow equations for a simple problem like reaching de nitions become more complex, but the assumed memory consistency model has a large e ect on the complexity of the problem. In this paper, we show a contrary example where traditional data ow analysis remains quite viable in a compiler for shared memory parallel programs in a structured distributed shared memory environment, in which a shared space of tuples is accessed by properly synchronized methods. We demonstrate how standard intraprocess data ow analysis remains viable over tuplespace communication statements, and present improvements to the precision of the analysis in the presence of these statements. We present a data ow framework to compute reaching de nitions across process boundaries, and a technique to improve the precision of this interprocess analysis. Lastly, we demonstrate some transformations enabled by reaching de nitions through tuplespace.