Can we apply accelerator-cores to control-intensive programs?

There is a trend towards using accelerators to increase performance and energy efficiency of general-purpose processors. So far, most accelerators have been build with HPC-applications in mind. A question that arises is how well can other applications benefit from these accelerators? In this paper, we discuss the acceleration of three benchmarks using the SPUs of a Cell-BE. We analyze the potential speedup given the inherent parallelism in the applications. While the potential speedup is significant in all benchmarks, the obtained speedup lags behind due to a mismatch between micro-architectural properties of the accelerators and the benchmark properties