Prototyping the DySER specialization architecture with OpenSPARC

This paper describes the prototype implementation of the DySER specialization architecture integrated into the OpenSPARC processor. The paper's description covers the hardware, compiler, and application tuning. The prototype system provides speedups up to 14× over OpenSPARC (geometric mean 5×). The architecture is more flexible than SIMD and GPU-based acceleration while supporting a more diverse set of workloads