An iteration space visualizer for polyhedral loop transformations in numerical programming

An iteration space visualizer is presented to analyze parallelism in loop nests including parallelism in tiled code of numerical programs. The tool visualizes exact data dependences available in arbitrarily nested loops as well as tiles generated with TRACO by means of the transitive closure of a loop nest dependence graph. Various graphical operations such as rotation, zooming, coloring and filtering allow for a detailed examination of dependences, iteration space slices, and shapes of generated tiles. The visualizer is a built-in TRACO module which collects results generated with TRACO and it is launched automatically when TRACO finishes code generation. The visualizer helps high-performance application developers discover parallelism available in loop nests and analyze tiled code produced by means of the polyhedral model.