Compiling C programs into threads

In recent years research in the area of parallel architectures and parallel languages has become more and more important. On one hand parallel languages offer the advantage to implement parallel algorithms efficiently, on the other hand most of the currently used software is still implemented in sequential languages. Rewriting this software will take a long time and is also error prone, so that the resources of parallel systems are actually not efficiently used. Especially numerical problems have been subject to automatic parallelisation. This work especially focussed on loops and also resulted in new language definitions. This paper describes methods for detecting coarse grain parallelism within scientific C-programs and presents basic algorithms for detecting and exploiting the parallelism and concludes with results obtained by analysis of traces of sequential programs, which reflect the possibilities of the approach shown here. (orig.)Available from TIB Hannover: RR 8747(93-4) / FIZ - Fachinformationszzentrum Karlsruhe / TIB - Technische InformationsbibliothekSIGLEDEGerman