On index set splitting

摘要: There are many algorithms for the space-time mapping of nested loops. Some them even make optimal choices within their framework. We propose a preprocessing phase in polytope model, which extends model and yields mappings whose schedule is, some cases, orders magnitude faster. These cases dependence graph has small irregularities. The basic idea is to split iteration domain loop nests into parts with regular structure apply existing these individually. work based on seminal more limited context parallelization at code level. elevate level (our model), increases its applicability by providing clearer wider range an acceptable analysis cost. Index set splitting one facet effort extend power enable generation competitive target code.