PUMA-V: Optimizing Parallel Code Performance Through Interactive Visualization

Performance optimization for parallel loop-oriented programs compromises between parallelism and locality. We present a visualization interface that allows programmers to assist the compiler in generating optimal code. It greatly improves the user's understanding of the transformations that took place and aids in making additional transformations in a visually intuitive way.

Visual analytics enables users to participate in machine-based optimization processes, contributing elements of human creativity, ingenuity, and expertise as well as commonsense knowledge. The benefits of visual analytics have been demonstrated in many domains, such as science, business, and medicine. In this paper, we take advantage of visual analytics to allow users to reason about parallel code generation. The visual tool we developed, PUMA-V, is especially useful for compiler developers who are looking to expose and patch weaknesses in the compiler's optimization pipeline, as well as general users who want to optimize their code but do not know how to best focus their efforts.