Aiding Parallel Programming with On-the-Fly Dependence Visualisation

Parallel programming is notoriously difficult. This becomes even more critical as multicore processors bring parallel computing into the mainstream. In order to ease the difficulty, tools have been designed that help the programmer with some aspects of parallelisation. Unfortunately, the programmer is mostly left along when it comes to the difficult task of dependence analysis among the subtasks to be executed concurrently. This paper presents a new visual tool that supports the programmer with the dependence analysis in loops. This is very useful in combination with an automatically parallelising compiler or when loops are parallelised with OpenMP. The tool displays on-the-fly the dependences between the statements of the loop nest on which the developer is currently working. To maximise the usefulness of the tool, it is unobtrusive, customisable and flexible, and based on dependence analysis theory. A prototype was implemented for the Eclipse IDE as a plug-in that seamlessly integrates into the normal development process. The evaluation of the tool, including an evaluation against cognitive dimensions, demonstrates the usability and usefulness of the tool.