The program dependence graph and its use in optimization

Abstract

In this paper we present an intermediate program representation, called the program dependence graph (PDG), that makes explicit both the data and control dependence5 for each operation in a program. Data dependences have been used to represent only the relevant data flow relationships of a program. Control dependence5 are introduced to analogously represent only the essential control flow relationships of a program. Control dependences are derived from the usual control flow graph. Many traditional optimizations operate more efficiently on the PDG. Since dependences in the PDG connect computationally related parts of the program, a single walk of these dependences is sufficient to perform many optimizations. The PDG allows transformations such as vectorization, that previously required special treatment of control dependence, to be performed in a manner that is uniform for both control and data dependences. Program transformations that require interaction of the two dependence types can also be easily handled with our representation. As an example, an incremental approach to modifying data dependences resulting from branch deletion or loop unrolling is introduced. The PDG supports incremental optimization, permitting transformations to be triggered by one another and applied only to affected dependences.

Applying aggressive propagation-based strategies for testing changes

“…The right of Figure 1 shows an almost complete Program Dependence Graph (PDG) [5] for this program. (In general, an interprocedural form of the PDG is needed [7].)…”

“…Node n 1 is control dependent [5] on node n 2 if the decision taken at n 2 determines whether n 1 is necessarily executed. Control dependencies are created by the presence of branches and other control decisions such as virtual calls.…”

“…Early techniques [3] [14] use controland data-dependence analysis [5] [20] to identify entities such as branches and definition-use pairs (du-pairs) that must be covered in the modified program because their execution behavior may be affected by the changes. These techniques, which we call coverage-based strategies, resemble typical testing criteria for programs in which entities of a certain type need to be covered.…”

Parametric and Sliced Causality

“…In particular, one can use off-the-shelf algorithms for classic [8] or for weak [17] control dependence. We chose to use the terminationsensitive control dependence (TSCD) introduced in [5] in our implementation of jPredictor [3].…”

Visualization of Program Dependence Graphs

“…The Java HotSpot TM server compiler [5] of Sun Microsystems uses a program dependence graph [2] as the intermediate data structure when compiling Java bytecodes to machine code. It applies optimizations such as global value numbering, conditional constant propagation, and loop transformations to produce faster code.…”