Adding Parallelism to Visual Data Flow Programs.

Philip Cox, Simon Gauvin, and Andrew Rau-Chaplin

Abstract: Programming in parallel is an error-prone and complex task compounded by the lack of tool support for both programming and debugging. Recent advances in compiler-directed shared memory APIs, such as OpenMP, have made shared-memory parallelism more widely accessible for users of traditional procedural languages: however, the mechanisms provided are difficult to use and error-prone. This paper examines the use
of visual notations for data flow programming to enable the creation of shared memory parallel programs. We present a model, arising from research on the ReactoGraph visual programming language, that allows code in a general class of visual data flow languages to be parallelized using visual annotations, and discuss the advantages this has over current textual methods.