Section: New Results

Visual Program Manipulation in the Polyhedral Model

Participants : Cédric Bastoul, Oleksandr Zinenko, Stéphane Huot.

While a plethora of libraries and frameworks focus on expressing parallelism, identifying and extracting it remains a challenging task. Automatic parallelization relies on imprecise heuristics resulting in cumbersome manual code analysis and transformation in case of underperformance. Alternatively, directive-based approaches often require transforming the program from scratch when a slightly modified version of an automatically-computed transformation would suffice. We propose an interactive visual approach building on the polyhedral model that (1) visualizes exact dependences and parallelism, (2) decomposes a complex automatically-computed transformation into simple steps for replay and easier modification, and (3) allows for directly manipulating the visual representation as a means of transforming the program with immediate feedback. User studies suggest that our visualization is understood by experts and non-experts alike, and that it may favor an exploratory approach to transformation. Finally, an eye-tracking study suggests that programmers may resort to visualizations instead of code if visualizations are clearly efficient for a given task.

This is a joint work with PARKAS team at Inria Paris (contact: Oleksandr Zinenko) and MJOLNIR team at Inria Lille (contact: Stéphane Huot), published in TACO [10].