Interference graph dataset for machine learning based register allocation

Register allocation is an important phase in compiler optimization. Often, its resolution involves graph coloring, which is an NP-complete problem. Because of their significance, numerous heuristics have been suggested for their resolution. Heuristic development is a complex process that requires specialized domain expertise. Recently, several machine learning based approaches have been proposed to solve compiler optimization problems. Still, due to the complexity of the problem and the lack of specialized datasets for training models applied to register allocation, few works researched this topic. The deficiency in sufficient adequate test cases is a recurring issue when working with register allocation, even beyond the scope of machine learning applications. In an effort to address this problem and facilitate forthcoming research in this domain, we present this dataset for training Machine Learning models focused on register allocation. The dataset consists of PBQP interference graphs from real C/C++ codes, and it is easily adaptable to various register allocation techniques.