Speculative Program Parallelization with Scalable and Decentralized Runtime Verification

摘要: Thread Level Speculation (TLS) is a dynamic code parallelization technique proposed to keep the software in pace with advances hardware, particular, automatically parallelize programs take advantage of multi-core processors. Being speculative, frameworks this type unavoidably rely on verification systems that are similar transactional memory, and require voluminous inter-thread communications or centralized registering performed memory accesses. The high degree communication against basic principles performance parallel computing, does not scale an increasing number processor cores, yields weak performance. Moreover, TLS often apply one unique strategy consisting slicing loop into several speculative threads. Such also since loops original serial necessarily also, it well-known schedule must promote data locality which crucial obtaining good This situation appeals scalable decentralized new strategies dynamically generate efficient resulting from advanced optimizing parallelizing transformations. transformations more complex system allows intra-thread iterations be reordered. In paper, we propose kind, based model built at runtime predicting linear behavior. part Apollo parallelizer adaptation for usage polyhedral model.