In this paper, we present automatic loop tiling and parallelization for the minimum cost polygon triangulation (MCPT) task. For this purpose, we use the authorial sourceto-source TRACO compiler. MCPT is a recursive algorithm encountering each subproblem many times in different branches of its recursion tree. The most intensive computing part is a triple nested polyhedral program loop nest filling a cost table using the MCPT recursive. First, the code is tiled by means of the transitive closure

more »

... f a dependence graph. TRACO allows for tiling of the innermost loop nest that is not possible by means of other closely related compilers. We tile only the two innermost loops and apply skewing to serialize the outermost one and parallelize the innermost ones. An experimental study carried out on multi-core computers demonstrates considerable speed-up of tiled code, which overcomes that obtained for code generated with the closely related PLuTo compiler based on the affine transformations framework.