Bioinformatics Specific Cell BE ISA Extensions

Among other traditionally important application domains, bioinformatics has been recently recognized as a challenging field. The need to perform lots of processing of huge amounts of data within reasonable times demands the use of high performance computing systems. Although this kind of high performance architectures already exists and can indeed provide some speedup, efficiency levels keep being low, both in terms of performance and power consumption. This is understandable since the traditional design of microprocessors has focused on more commercially attractive application domains like multimedia and networking. With the arrival of the multicore era, researchers have already started mapping bioinformatics applications onto platforms like Cell BE. This type of heterogeneous multicore chips have been shown to be capable of achieving significant speedups by exploiting threadlevel and data-level parallelism at the same time. However, it has also been demonstrated that architectures like Cell BE have shortcomings that limit the performance efficiency when targeting bioinformatics workloads. Based on a previous work, we have augmented the instruction-set of the Cell BE SPU, with the aim of accelerating bioinformatics applications. We started looking at the possible improvements of the computational parts of a representative application of the field, i.e. ClustalW. Four arithmetic instructions have been tested with the CellSim simulator, achieving an accumulated acceleration of almost 14%.