Parallel pattern detection for architectural improvements
Jason A. Poovey, Brian P. Railing, Thomas M. Conte · USENIX conference on Hot topics in parallelism · 2011
With the shift in general purpose computing to increasingly parallel architectures comes a need for clever architectures to achieve high parallelism on previously sequential or poorly parallelized code. In order to fully utilize the many-core systems of the present and future, a shift must occur in architecture design philosophy to understanding how the parallel programming process affects design decisions. Parallel patterns provide a way to create parallel code for a wide variety of algorithms. Additionally they provide a convenient classification mechanism that is both understandable to programmers and that exhibit similar behaviors that can be architecturally exploited. In this work we explore the capabilities of pattern driven dynamic architectures as well as detection mechanisms useful for dynamic and static parallel pattern recognition.