The Spectrum Of Thread Implementations On Hybrid Multithreaded Architectures

OF DISSERTATION THE SPECTRUM OF THREAD IMPLEMENTATIONS ON HYBRID MULTITHREADED ARCHITECTURES The proliferation of multithreaded architectures indicates a desire to combine the best of the dataflow and von Neumann models. These architectures have the capability of masking memory latencies using the dataflow model as well as the capability of exploiting locality, both temporal and spatial, a la von Neumann. This is accomplished by increasing granularity from instruction level to thread level. Increased granularity results in reducing the synchronization required per instruction and introduces locality where none was present. Applicative and functional languages expose parallelism with no intervention by the programmer and without expensive data dependence analysis, apart from array analysis. With the combination of applicative languages, hybrid multithreaded architectures and capable compilers, a possible platform for scalable high performance computing is created. The challenge is to de..