Avoiding unnecessary remappings of data arrays at run-time is a major
means for improving the performance of data-parallel programs.
Recently, we presented a novel, aggressive intraprocedural
approach achieving this by eliminating partially redundant and
partially dead distribution assignments. Here, we focus on the
analogies and differences showing up when extending this approach of
distribution assignment placement (DAP) interprocedurally.
Above all, we show that (1) the optimality results applying
intraprocedurally are interprocedurally in general impossible, and
(2) the strategies leading to optimal results intraprocedurally can
cause dramatic disimprovements interprocedurally. Nonetheless, we
demonstrate how to arrive at a powerful and flexible
approach for interprocedural DAP (IDAP) leading as its
intraprocedural counterpart to a hierarchy of IDAP-algorithms of
varying power and efficiency. First empirical results underline its
effectivity and practical relevance.