Classification of Visualized Data Dependence

摘要

Recently, as the use of parallel computers becomes popular in the field of scientific calculation, effective converting methods from conventional sequential programs into corresponding parallel ones are required. For effective conversions, however, vast amounts of skills and knowledge for parallel programming are needed for programmers. In this paper, we propose the intro-ducion of a 3D visualization system called NaraView as a programming environment for the effective paralleliza-tion. NaraView analyzes data dependences in a given program code, and visualizes their relations. By the visualization of program codes, we can recognize typical patterns in the shape of visualized data dependences, and implement corresponding parallelizing methods easily. We show effective parallelizing methods which have such typical patterns in loop structures of concrete program codes.