Abstract: "Program restructuring is a subject that has been studied for many years by a few researchers, and which has attracted great attention in recent years [AlCo72], [AlKe82], [Bane76], [Bane79], [Cytr84], [Kuck78], [Padu79], [PaWo86], [Poly86], [Wolf82]. Many vectorizing compilers have be...

Ausführliche Beschreibung

Bibliographische Detailangaben

1. Verfasser:	Polychronopoulos, Constantine D. 1958- (VerfasserIn)
Format:	Buch
Sprache:	English
Veröffentlicht:	Urbana, Ill. 1987
Schriftenreihe:	Center for Supercomputing Research and Development <Urbana, Ill.>: CSRD report 667
Schlagworte:	Compiling (Electronic computers) Parallel processing (Electronic computers) Transformations (Mathematics)
Zusammenfassung:	Abstract: "Program restructuring is a subject that has been studied for many years by a few researchers, and which has attracted great attention in recent years [AlCo72], [AlKe82], [Bane76], [Bane79], [Cytr84], [Kuck78], [Padu79], [PaWo86], [Poly86], [Wolf82]. Many vectorizing compilers have been developed in the past ten years [KKLW80], [Alli85], [Kenn80], [Brod81]. However, restructuring for parallel machines is still a relatively new subject. Some of the parallelizing compilers under development are the University of Illinois Parafrase II, the Rice University PFC, and the IBM PTRAN. In this paper we discuss two new techniques for transforming serial programs into a parallel form suitable for execution on parallel processor systems These schemes can be implemented in a restructuring compiler with minimal effort, and for certain types of serial loops they can obtain significant improvement. More specifically, we discuss in detail cycle shrinking, a compiler transformation that can partially parallelize certain types of serial loops. Three versions of this transformation (based on different uses of dependence distance information) are presented and compared. The second major subject of this paper is run-time dependence analysis, through compiler-inserted bookkeeping and control statements. Loops with unstructured parallelism that cannot benefit from existing optimizations, can be parallelized through run-time dependence checking The rest of the material is organized as follows. Section 2 gives a short introductions to program dependences and defines the basic concepts. Section 3 discusses cycle shrinking in detail; Section 3.1 covers singly nested loops, while Section 3.2 discusses the application of cycle shrinking to complex nested loops. Run-time dependence checking and the RDC transformation are presented in Section 4. The conclusion is given in Section 5.
Beschreibung:	30 S.