Scratchpad Allocation for Data Aggregates in Superperfect Graphs

Existing methods place data or code in scratchpad memory, i.e., SPM by either relying on heuristics or resorting to integer programming or mapping it to a graph coloring problem. In this work, the SPM allocation problem is formulated as an interval coloring problem. The key observation is that in many embedded applications, arrays (including structs as a special case) are often related in the following way: For any two arrays, their live ranges are often such that one is either disjoint from or contains the other. As a result, array interference graphs are often superperfect graphs and optimal interval colorings for such array interference graphs are possible. This has led to the development of two new SPM allocation algorithms. While differing in whether live range splits and spills are done sequentially or together, both algorithms place arrays in SPM based on examining the cliques in an interference graph. In both cases, we guarantee optimally that all arrays in an interference graph can be placed in SPM if its size is no smaller than the clique number of the graph. In the case that the SPM is not large enough, we rely on heuristics to split or spill a live range until the graph is colorable. Our experiment results using embedded benchmarks show that our algorithms can outperform graph coloring when their interference graphs are superperfect or nearly so although graph coloring is admittedly more general and may also be effective to applications with arbitrary interference graphs